mirrors/opencv_contrib
1
0
Fork 0
Code Pull Requests Packages Releases Activity

vendor: OpenCV 5.0.0 snapshot at 755e50675d97db9b7d449d8bd6b09888646f6c6e

Browse Source
This commit is contained in:
Gitea Mirror Bot
2026-08-22 00:11:13 +08:00
commit 12022378a3
3872 changed files with 2513409 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/cudalegacy/CMakeLists.txt
+21
View File
@@ -0,0 +1,21 @@
if(NOT HAVE_CUDA)
ocv_module_disable(cudalegacy)
endif()
set(the_description "CUDA-accelerated Computer Vision (legacy)")
ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4127 /wd4130 /wd4324 /wd4512 /wd4310 -Wundef -Wmissing-declarations -Wuninitialized -Wshadow -Wdeprecated-declarations -Wstrict-aliasing -Wtautological-compare)
ocv_add_module(cudalegacy opencv_core opencv_geometry opencv_video OPTIONAL opencv_objdetect opencv_xobjdetect opencv_imgproc opencv_geometry opencv_stereo opencv_calib opencv_cudaarithm opencv_cudafilters opencv_cudaimgproc)
ocv_module_include_directories()
ocv_glob_module_sources()
set(extra_libs "")
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
list(APPEND extra_libs CUDA::cudart${CUDA_LIB_EXT})
endif()
ocv_create_module(${extra_libs})
ocv_add_accuracy_tests(${extra_libs})
ocv_add_perf_tests(${extra_libs})
modules/cudalegacy/include/opencv2/cudalegacy.hpp
+293
View File
@@ -0,0 +1,293 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef OPENCV_CUDALEGACY_HPP
#define OPENCV_CUDALEGACY_HPP
#include "opencv2/core/cuda.hpp"
#if defined (HAVE_CUDA) && !defined (CUDA_DISABLER)
#include "opencv2/cudalegacy/NCV.hpp"
#include "opencv2/cudalegacy/NPP_staging.hpp"
#include "opencv2/cudalegacy/NCVPyramid.hpp"
#include "opencv2/cudalegacy/NCVHaarObjectDetection.hpp"
#include "opencv2/cudalegacy/NCVBroxOpticalFlow.hpp"
#endif
#include "opencv2/video/background_segm.hpp"
/**
@addtogroup cuda
@{
@defgroup cudalegacy Legacy support
@}
*/
namespace cv { namespace cuda {
//! @addtogroup cudalegacy
//! @{
//
// ImagePyramid
//
class CV_EXPORTS ImagePyramid : public Algorithm
{
public:
virtual void getLayer(OutputArray outImg, Size outRoi, Stream& stream = Stream::Null()) const = 0;
};
CV_EXPORTS Ptr<ImagePyramid> createImagePyramid(InputArray img, int nLayers = -1, Stream& stream = Stream::Null());
//
// GMG
//
/** @brief Background/Foreground Segmentation Algorithm.
The class discriminates between foreground and background pixels by building and maintaining a model
of the background. Any pixel which does not fit this model is then deemed to be foreground. The
class implements algorithm described in @cite Gold2012 .
*/
class CV_EXPORTS BackgroundSubtractorGMG : public cv::BackgroundSubtractor
{
public:
using cv::BackgroundSubtractor::apply;
virtual void apply(InputArray image, OutputArray fgmask, double learningRate, Stream& stream) = 0;
virtual void apply(InputArray image, InputArray knownForegroundMask, OutputArray fgmask, double learningRate, Stream& stream) = 0;
virtual int getMaxFeatures() const = 0;
virtual void setMaxFeatures(int maxFeatures) = 0;
virtual double getDefaultLearningRate() const = 0;
virtual void setDefaultLearningRate(double lr) = 0;
virtual int getNumFrames() const = 0;
virtual void setNumFrames(int nframes) = 0;
virtual int getQuantizationLevels() const = 0;
virtual void setQuantizationLevels(int nlevels) = 0;
virtual double getBackgroundPrior() const = 0;
virtual void setBackgroundPrior(double bgprior) = 0;
virtual int getSmoothingRadius() const = 0;
virtual void setSmoothingRadius(int radius) = 0;
virtual double getDecisionThreshold() const = 0;
virtual void setDecisionThreshold(double thresh) = 0;
virtual bool getUpdateBackgroundModel() const = 0;
virtual void setUpdateBackgroundModel(bool update) = 0;
virtual double getMinVal() const = 0;
virtual void setMinVal(double val) = 0;
virtual double getMaxVal() const = 0;
virtual void setMaxVal(double val) = 0;
};
/** @brief Creates GMG Background Subtractor
@param initializationFrames Number of frames of video to use to initialize histograms.
@param decisionThreshold Value above which pixel is determined to be FG.
*/
CV_EXPORTS Ptr<cuda::BackgroundSubtractorGMG>
createBackgroundSubtractorGMG(int initializationFrames = 120, double decisionThreshold = 0.8);
//
// FGD
//
/** @brief The class discriminates between foreground and background pixels by building and maintaining a model
of the background.
Any pixel which does not fit this model is then deemed to be foreground. The class implements
algorithm described in @cite FGD2003 .
@sa BackgroundSubtractor
*/
class CV_EXPORTS BackgroundSubtractorFGD : public cv::BackgroundSubtractor
{
public:
/** @brief Returns the output foreground regions calculated by findContours.
@param foreground_regions Output array (CPU memory).
*/
virtual void getForegroundRegions(OutputArrayOfArrays foreground_regions) = 0;
};
struct CV_EXPORTS FGDParams
{
int Lc; //!< Quantized levels per 'color' component. Power of two, typically 32, 64 or 128.
int N1c; //!< Number of color vectors used to model normal background color variation at a given pixel.
int N2c; //!< Number of color vectors retained at given pixel. Must be > N1c, typically ~ 5/3 of N1c.
//!< Used to allow the first N1c vectors to adapt over time to changing background.
int Lcc; //!< Quantized levels per 'color co-occurrence' component. Power of two, typically 16, 32 or 64.
int N1cc; //!< Number of color co-occurrence vectors used to model normal background color variation at a given pixel.
int N2cc; //!< Number of color co-occurrence vectors retained at given pixel. Must be > N1cc, typically ~ 5/3 of N1cc.
//!< Used to allow the first N1cc vectors to adapt over time to changing background.
bool is_obj_without_holes; //!< If TRUE we ignore holes within foreground blobs. Defaults to TRUE.
int perform_morphing; //!< Number of erode-dilate-erode foreground-blob cleanup iterations.
//!< These erase one-pixel junk blobs and merge almost-touching blobs. Default value is 1.
float alpha1; //!< How quickly we forget old background pixel values seen. Typically set to 0.1.
float alpha2; //!< "Controls speed of feature learning". Depends on T. Typical value circa 0.005.
float alpha3; //!< Alternate to alpha2, used (e.g.) for quicker initial convergence. Typical value 0.1.
float delta; //!< Affects color and color co-occurrence quantization, typically set to 2.
float T; //!< A percentage value which determines when new features can be recognized as new background. (Typically 0.9).
float minArea; //!< Discard foreground blobs whose bounding box is smaller than this threshold.
//! default Params
FGDParams();
};
/** @brief Creates FGD Background Subtractor
@param params Algorithm's parameters. See @cite FGD2003 for explanation.
*/
CV_EXPORTS Ptr<cuda::BackgroundSubtractorFGD>
createBackgroundSubtractorFGD(const FGDParams& params = FGDParams());
//
// Optical flow
//
//! Calculates optical flow for 2 images using block matching algorithm */
CV_EXPORTS void calcOpticalFlowBM(const GpuMat& prev, const GpuMat& curr,
Size block_size, Size shift_size, Size max_range, bool use_previous,
GpuMat& velx, GpuMat& vely, GpuMat& buf,
Stream& stream = Stream::Null());
class CV_EXPORTS FastOpticalFlowBM
{
public:
void operator ()(const GpuMat& I0, const GpuMat& I1, GpuMat& flowx, GpuMat& flowy, int search_window = 21, int block_window = 7, Stream& s = Stream::Null());
private:
GpuMat buffer;
GpuMat extended_I0;
GpuMat extended_I1;
};
/** @brief Interpolates frames (images) using provided optical flow (displacement field).
@param frame0 First frame (32-bit floating point images, single channel).
@param frame1 Second frame. Must have the same type and size as frame0 .
@param fu Forward horizontal displacement.
@param fv Forward vertical displacement.
@param bu Backward horizontal displacement.
@param bv Backward vertical displacement.
@param pos New frame position.
@param newFrame Output image.
@param buf Temporary buffer, will have width x 6\*height size, CV_32FC1 type and contain 6
GpuMat: occlusion masks for first frame, occlusion masks for second, interpolated forward
horizontal flow, interpolated forward vertical flow, interpolated backward horizontal flow,
interpolated backward vertical flow.
@param stream Stream for the asynchronous version.
*/
CV_EXPORTS void interpolateFrames(const GpuMat& frame0, const GpuMat& frame1,
const GpuMat& fu, const GpuMat& fv,
const GpuMat& bu, const GpuMat& bv,
float pos, GpuMat& newFrame, GpuMat& buf,
Stream& stream = Stream::Null());
CV_EXPORTS void createOpticalFlowNeedleMap(const GpuMat& u, const GpuMat& v, GpuMat& vertex, GpuMat& colors);
//
// Labeling
//
//!performs labeling via graph cuts of a 2D regular 4-connected graph.
CV_EXPORTS void graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& bottom, GpuMat& labels,
GpuMat& buf, Stream& stream = Stream::Null());
//!performs labeling via graph cuts of a 2D regular 8-connected graph.
CV_EXPORTS void graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& topLeft, GpuMat& topRight,
GpuMat& bottom, GpuMat& bottomLeft, GpuMat& bottomRight,
GpuMat& labels,
GpuMat& buf, Stream& stream = Stream::Null());
//! compute mask for Generalized Flood fill componetns labeling.
CV_EXPORTS void connectivityMask(const GpuMat& image, GpuMat& mask, const cv::Scalar& lo, const cv::Scalar& hi, Stream& stream = Stream::Null());
//! performs connected componnents labeling.
CV_EXPORTS void labelComponents(const GpuMat& mask, GpuMat& components, int flags = 0, Stream& stream = Stream::Null());
//
// Calib3d
//
CV_EXPORTS void transformPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec,
GpuMat& dst, Stream& stream = Stream::Null());
CV_EXPORTS void projectPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec,
const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst,
Stream& stream = Stream::Null());
/** @brief Finds the object pose from 3D-2D point correspondences.
@param object Single-row matrix of object points.
@param image Single-row matrix of image points.
@param camera_mat 3x3 matrix of intrinsic camera parameters.
@param dist_coef Distortion coefficients. See undistortPoints for details.
@param rvec Output 3D rotation vector.
@param tvec Output 3D translation vector.
@param use_extrinsic_guess Flag to indicate that the function must use rvec and tvec as an
initial transformation guess. It is not supported for now.
@param num_iters Maximum number of RANSAC iterations.
@param max_dist Euclidean distance threshold to detect whether point is inlier or not.
@param min_inlier_count Flag to indicate that the function must stop if greater or equal number
of inliers is achieved. It is not supported for now.
@param inliers Output vector of inlier indices.
*/
CV_EXPORTS void solvePnPRansac(const Mat& object, const Mat& image, const Mat& camera_mat,
const Mat& dist_coef, Mat& rvec, Mat& tvec, bool use_extrinsic_guess=false,
int num_iters=100, float max_dist=8.0, int min_inlier_count=100,
std::vector<int>* inliers=NULL);
//! @}
}}
#endif /* OPENCV_CUDALEGACY_HPP */
modules/cudalegacy/include/opencv2/cudalegacy/NCV.hpp
+1032
View File
File diff suppressed because it is too large Load Diff
modules/cudalegacy/include/opencv2/cudalegacy/NCVBroxOpticalFlow.hpp
+110
View File
@@ -0,0 +1,110 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
////////////////////////////////////////////////////////////////////////////////
//
// NVIDIA CUDA implementation of Brox et al Optical Flow algorithm
//
// Algorithm is explained in the original paper:
// T. Brox, A. Bruhn, N. Papenberg, J. Weickert:
// High accuracy optical flow estimation based on a theory for warping.
// ECCV 2004.
//
// Implementation by Mikhail Smirnov
// email: msmirnov@nvidia.com, devsupport@nvidia.com
//
// Credits for help with the code to:
// Alexey Mendelenko, Anton Obukhov, and Alexander Kharlamov.
//
////////////////////////////////////////////////////////////////////////////////
#ifndef _ncv_optical_flow_h_
#define _ncv_optical_flow_h_
#include "opencv2/cudalegacy/NCV.hpp"
//! @addtogroup cudalegacy
//! @{
/// \brief Model and solver parameters
struct NCVBroxOpticalFlowDescriptor
{
/// flow smoothness
Ncv32f alpha;
/// gradient constancy importance
Ncv32f gamma;
/// pyramid scale factor
Ncv32f scale_factor;
/// number of lagged non-linearity iterations (inner loop)
Ncv32u number_of_inner_iterations;
/// number of warping iterations (number of pyramid levels)
Ncv32u number_of_outer_iterations;
/// number of linear system solver iterations
Ncv32u number_of_solver_iterations;
};
/////////////////////////////////////////////////////////////////////////////////////////
/// \brief Compute optical flow
///
/// Based on method by Brox et al [2004]
/// \param [in] desc model and solver parameters
/// \param [in] gpu_mem_allocator GPU memory allocator
/// \param [in] frame0 source frame
/// \param [in] frame1 frame to track
/// \param [out] u flow horizontal component (along \b x axis)
/// \param [out] v flow vertical component (along \b y axis)
/// \param stream
/// \return computation status
/////////////////////////////////////////////////////////////////////////////////////////
CV_EXPORTS
NCVStatus NCVBroxOpticalFlow(const NCVBroxOpticalFlowDescriptor desc,
INCVMemAllocator &gpu_mem_allocator,
const NCVMatrix<Ncv32f> &frame0,
const NCVMatrix<Ncv32f> &frame1,
NCVMatrix<Ncv32f> &u,
NCVMatrix<Ncv32f> &v,
cudaStream_t stream);
//! @}
#endif
modules/cudalegacy/include/opencv2/cudalegacy/NCVHaarObjectDetection.hpp
+464
View File
@@ -0,0 +1,464 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
////////////////////////////////////////////////////////////////////////////////
//
// NVIDIA CUDA implementation of Viola-Jones Object Detection Framework
//
// The algorithm and code are explained in the upcoming GPU Computing Gems
// chapter in detail:
//
// Anton Obukhov, "Haar Classifiers for Object Detection with CUDA"
// PDF URL placeholder
// email: aobukhov@nvidia.com, devsupport@nvidia.com
//
// Credits for help with the code to:
// Alexey Mendelenko, Cyril Crassin, and Mikhail Smirnov.
//
////////////////////////////////////////////////////////////////////////////////
#ifndef _ncvhaarobjectdetection_hpp_
#define _ncvhaarobjectdetection_hpp_
#include "opencv2/cudalegacy/NCV.hpp"
//! @addtogroup cudalegacy
//! @{
//==============================================================================
//
// Guaranteed size cross-platform classifier structures
//
//==============================================================================
#if defined __GNUC__ && (__GNUC__*100 + __GNUC_MINOR__ > 204)
typedef Ncv32f __attribute__((__may_alias__)) Ncv32f_a;
#else
typedef Ncv32f Ncv32f_a;
#endif
struct HaarFeature64
{
uint2 _ui2;
#define HaarFeature64_CreateCheck_MaxRectField 0xFF
__host__ NCVStatus setRect(Ncv32u rectX, Ncv32u rectY, Ncv32u rectWidth, Ncv32u rectHeight, Ncv32u /*clsWidth*/, Ncv32u /*clsHeight*/)
{
ncvAssertReturn(rectWidth <= HaarFeature64_CreateCheck_MaxRectField && rectHeight <= HaarFeature64_CreateCheck_MaxRectField, NCV_HAAR_TOO_LARGE_FEATURES);
NcvRect8u* tmpRect = (NcvRect8u*)&this->_ui2.x;
tmpRect->x = (Ncv8u)rectX;
tmpRect->y = (Ncv8u)rectY;
tmpRect->width = (Ncv8u)rectWidth;
tmpRect->height = (Ncv8u)rectHeight;
return NCV_SUCCESS;
}
__host__ NCVStatus setWeight(Ncv32f weight)
{
((Ncv32f_a*)&(this->_ui2.y))[0] = weight;
return NCV_SUCCESS;
}
__device__ __host__ void getRect(Ncv32u *rectX, Ncv32u *rectY, Ncv32u *rectWidth, Ncv32u *rectHeight)
{
NcvRect8u* tmpRect = (NcvRect8u*)&this->_ui2.x;
*rectX = tmpRect->x;
*rectY = tmpRect->y;
*rectWidth = tmpRect->width;
*rectHeight = tmpRect->height;
}
__device__ __host__ Ncv32f getWeight(void)
{
return *(Ncv32f_a*)(&this->_ui2.y);
}
};
struct HaarFeatureDescriptor32
{
private:
#define HaarFeatureDescriptor32_Interpret_MaskFlagTilted 0x80000000
#define HaarFeatureDescriptor32_Interpret_MaskFlagLeftNodeLeaf 0x40000000
#define HaarFeatureDescriptor32_Interpret_MaskFlagRightNodeLeaf 0x20000000
#define HaarFeatureDescriptor32_CreateCheck_MaxNumFeatures 0x1F
#define HaarFeatureDescriptor32_NumFeatures_Shift 24
#define HaarFeatureDescriptor32_CreateCheck_MaxFeatureOffset 0x00FFFFFF
Ncv32u desc;
public:
__host__ NCVStatus create(NcvBool bTilted, NcvBool bLeftLeaf, NcvBool bRightLeaf,
Ncv32u numFeatures, Ncv32u offsetFeatures)
{
if (numFeatures > HaarFeatureDescriptor32_CreateCheck_MaxNumFeatures)
{
return NCV_HAAR_TOO_MANY_FEATURES_IN_CLASSIFIER;
}
if (offsetFeatures > HaarFeatureDescriptor32_CreateCheck_MaxFeatureOffset)
{
return NCV_HAAR_TOO_MANY_FEATURES_IN_CASCADE;
}
this->desc = 0;
this->desc |= (bTilted ? HaarFeatureDescriptor32_Interpret_MaskFlagTilted : 0);
this->desc |= (bLeftLeaf ? HaarFeatureDescriptor32_Interpret_MaskFlagLeftNodeLeaf : 0);
this->desc |= (bRightLeaf ? HaarFeatureDescriptor32_Interpret_MaskFlagRightNodeLeaf : 0);
this->desc |= (numFeatures << HaarFeatureDescriptor32_NumFeatures_Shift);
this->desc |= offsetFeatures;
return NCV_SUCCESS;
}
__device__ __host__ NcvBool isTilted(void)
{
return (this->desc & HaarFeatureDescriptor32_Interpret_MaskFlagTilted) != 0;
}
__device__ __host__ NcvBool isLeftNodeLeaf(void)
{
return (this->desc & HaarFeatureDescriptor32_Interpret_MaskFlagLeftNodeLeaf) != 0;
}
__device__ __host__ NcvBool isRightNodeLeaf(void)
{
return (this->desc & HaarFeatureDescriptor32_Interpret_MaskFlagRightNodeLeaf) != 0;
}
__device__ __host__ Ncv32u getNumFeatures(void)
{
return (this->desc >> HaarFeatureDescriptor32_NumFeatures_Shift) & HaarFeatureDescriptor32_CreateCheck_MaxNumFeatures;
}
__device__ __host__ Ncv32u getFeaturesOffset(void)
{
return this->desc & HaarFeatureDescriptor32_CreateCheck_MaxFeatureOffset;
}
};
struct HaarClassifierNodeDescriptor32
{
uint1 _ui1;
__host__ NCVStatus create(Ncv32f leafValue)
{
*(Ncv32f_a *)&this->_ui1 = leafValue;
return NCV_SUCCESS;
}
__host__ NCVStatus create(Ncv32u offsetHaarClassifierNode)
{
this->_ui1.x = offsetHaarClassifierNode;
return NCV_SUCCESS;
}
__host__ Ncv32f getLeafValueHost(void)
{
return *(Ncv32f_a *)&this->_ui1.x;
}
#ifdef __CUDACC__
__device__ Ncv32f getLeafValue(void)
{
return __int_as_float(this->_ui1.x);
}
#endif
__device__ __host__ Ncv32u getNextNodeOffset(void)
{
return this->_ui1.x;
}
};
#if defined __GNUC__ && (__GNUC__*100 + __GNUC_MINOR__ > 204)
typedef Ncv32u __attribute__((__may_alias__)) Ncv32u_a;
#else
typedef Ncv32u Ncv32u_a;
#endif
struct HaarClassifierNode128
{
uint4 _ui4;
__host__ NCVStatus setFeatureDesc(HaarFeatureDescriptor32 f)
{
this->_ui4.x = *(Ncv32u *)&f;
return NCV_SUCCESS;
}
__host__ NCVStatus setThreshold(Ncv32f t)
{
this->_ui4.y = *(Ncv32u_a *)&t;
return NCV_SUCCESS;
}
__host__ NCVStatus setLeftNodeDesc(HaarClassifierNodeDescriptor32 nl)
{
this->_ui4.z = *(Ncv32u_a *)&nl;
return NCV_SUCCESS;
}
__host__ NCVStatus setRightNodeDesc(HaarClassifierNodeDescriptor32 nr)
{
this->_ui4.w = *(Ncv32u_a *)&nr;
return NCV_SUCCESS;
}
__host__ __device__ HaarFeatureDescriptor32 getFeatureDesc(void)
{
return *(HaarFeatureDescriptor32 *)&this->_ui4.x;
}
__host__ __device__ Ncv32f getThreshold(void)
{
return *(Ncv32f_a*)&this->_ui4.y;
}
__host__ __device__ HaarClassifierNodeDescriptor32 getLeftNodeDesc(void)
{
return *(HaarClassifierNodeDescriptor32 *)&this->_ui4.z;
}
__host__ __device__ HaarClassifierNodeDescriptor32 getRightNodeDesc(void)
{
return *(HaarClassifierNodeDescriptor32 *)&this->_ui4.w;
}
};
struct HaarStage64
{
#define HaarStage64_Interpret_MaskRootNodes 0x0000FFFF
#define HaarStage64_Interpret_MaskRootNodeOffset 0xFFFF0000
#define HaarStage64_Interpret_ShiftRootNodeOffset 16
uint2 _ui2;
__host__ NCVStatus setStageThreshold(Ncv32f t)
{
this->_ui2.x = *(Ncv32u_a *)&t;
return NCV_SUCCESS;
}
__host__ NCVStatus setStartClassifierRootNodeOffset(Ncv32u val)
{
if (val > (HaarStage64_Interpret_MaskRootNodeOffset >> HaarStage64_Interpret_ShiftRootNodeOffset))
{
return NCV_HAAR_XML_LOADING_EXCEPTION;
}
this->_ui2.y = (val << HaarStage64_Interpret_ShiftRootNodeOffset) | (this->_ui2.y & HaarStage64_Interpret_MaskRootNodes);
return NCV_SUCCESS;
}
__host__ NCVStatus setNumClassifierRootNodes(Ncv32u val)
{
if (val > HaarStage64_Interpret_MaskRootNodes)
{
return NCV_HAAR_XML_LOADING_EXCEPTION;
}
this->_ui2.y = val | (this->_ui2.y & HaarStage64_Interpret_MaskRootNodeOffset);
return NCV_SUCCESS;
}
__host__ __device__ Ncv32f getStageThreshold(void)
{
return *(Ncv32f_a*)&this->_ui2.x;
}
__host__ __device__ Ncv32u getStartClassifierRootNodeOffset(void)
{
return (this->_ui2.y >> HaarStage64_Interpret_ShiftRootNodeOffset);
}
__host__ __device__ Ncv32u getNumClassifierRootNodes(void)
{
return (this->_ui2.y & HaarStage64_Interpret_MaskRootNodes);
}
};
NCV_CT_ASSERT(sizeof(HaarFeature64) == 8);
NCV_CT_ASSERT(sizeof(HaarFeatureDescriptor32) == 4);
NCV_CT_ASSERT(sizeof(HaarClassifierNodeDescriptor32) == 4);
NCV_CT_ASSERT(sizeof(HaarClassifierNode128) == 16);
NCV_CT_ASSERT(sizeof(HaarStage64) == 8);
//==============================================================================
//
// Classifier cascade descriptor
//
//==============================================================================
struct HaarClassifierCascadeDescriptor
{
Ncv32u NumStages;
Ncv32u NumClassifierRootNodes;
Ncv32u NumClassifierTotalNodes;
Ncv32u NumFeatures;
NcvSize32u ClassifierSize;
NcvBool bNeedsTiltedII;
NcvBool bHasStumpsOnly;
};
//==============================================================================
//
// Functional interface
//
//==============================================================================
enum
{
NCVPipeObjDet_Default = 0x000,
NCVPipeObjDet_UseFairImageScaling = 0x001,
NCVPipeObjDet_FindLargestObject = 0x002,
NCVPipeObjDet_VisualizeInPlace = 0x004,
};
CV_EXPORTS NCVStatus ncvDetectObjectsMultiScale_device(NCVMatrix<Ncv8u> &d_srcImg,
NcvSize32u srcRoi,
NCVVector<NcvRect32u> &d_dstRects,
Ncv32u &dstNumRects,
HaarClassifierCascadeDescriptor &haar,
NCVVector<HaarStage64> &h_HaarStages,
NCVVector<HaarStage64> &d_HaarStages,
NCVVector<HaarClassifierNode128> &d_HaarNodes,
NCVVector<HaarFeature64> &d_HaarFeatures,
NcvSize32u minObjSize,
Ncv32u minNeighbors, //default 4
Ncv32f scaleStep, //default 1.2f
Ncv32u pixelStep, //default 1
Ncv32u flags, //default NCVPipeObjDet_Default
INCVMemAllocator &gpuAllocator,
INCVMemAllocator &cpuAllocator,
cudaDeviceProp &devProp,
cudaStream_t cuStream);
#define OBJDET_MASK_ELEMENT_INVALID_32U 0xFFFFFFFF
#define HAAR_STDDEV_BORDER 1
CV_EXPORTS NCVStatus ncvApplyHaarClassifierCascade_device(NCVMatrix<Ncv32u> &d_integralImage,
NCVMatrix<Ncv32f> &d_weights,
NCVMatrixAlloc<Ncv32u> &d_pixelMask,
Ncv32u &numDetections,
HaarClassifierCascadeDescriptor &haar,
NCVVector<HaarStage64> &h_HaarStages,
NCVVector<HaarStage64> &d_HaarStages,
NCVVector<HaarClassifierNode128> &d_HaarNodes,
NCVVector<HaarFeature64> &d_HaarFeatures,
NcvBool bMaskElements,
NcvSize32u anchorsRoi,
Ncv32u pixelStep,
Ncv32f scaleArea,
INCVMemAllocator &gpuAllocator,
INCVMemAllocator &cpuAllocator,
cudaDeviceProp &devProp,
cudaStream_t cuStream);
CV_EXPORTS NCVStatus ncvApplyHaarClassifierCascade_host(NCVMatrix<Ncv32u> &h_integralImage,
NCVMatrix<Ncv32f> &h_weights,
NCVMatrixAlloc<Ncv32u> &h_pixelMask,
Ncv32u &numDetections,
HaarClassifierCascadeDescriptor &haar,
NCVVector<HaarStage64> &h_HaarStages,
NCVVector<HaarClassifierNode128> &h_HaarNodes,
NCVVector<HaarFeature64> &h_HaarFeatures,
NcvBool bMaskElements,
NcvSize32u anchorsRoi,
Ncv32u pixelStep,
Ncv32f scaleArea);
#define RECT_SIMILARITY_PROPORTION 0.2f
CV_EXPORTS NCVStatus ncvGrowDetectionsVector_device(NCVVector<Ncv32u> &pixelMask,
Ncv32u numPixelMaskDetections,
NCVVector<NcvRect32u> &hypotheses,
Ncv32u &totalDetections,
Ncv32u totalMaxDetections,
Ncv32u rectWidth,
Ncv32u rectHeight,
Ncv32f curScale,
cudaStream_t cuStream);
CV_EXPORTS NCVStatus ncvGrowDetectionsVector_host(NCVVector<Ncv32u> &pixelMask,
Ncv32u numPixelMaskDetections,
NCVVector<NcvRect32u> &hypotheses,
Ncv32u &totalDetections,
Ncv32u totalMaxDetections,
Ncv32u rectWidth,
Ncv32u rectHeight,
Ncv32f curScale);
CV_EXPORTS NCVStatus ncvHaarGetClassifierSize(const cv::String &filename, Ncv32u &numStages,
Ncv32u &numNodes, Ncv32u &numFeatures);
CV_EXPORTS NCVStatus ncvHaarLoadFromFile_host(const cv::String &filename,
HaarClassifierCascadeDescriptor &haar,
NCVVector<HaarStage64> &h_HaarStages,
NCVVector<HaarClassifierNode128> &h_HaarNodes,
NCVVector<HaarFeature64> &h_HaarFeatures);
CV_EXPORTS NCVStatus ncvHaarStoreNVBIN_host(const cv::String &filename,
HaarClassifierCascadeDescriptor haar,
NCVVector<HaarStage64> &h_HaarStages,
NCVVector<HaarClassifierNode128> &h_HaarNodes,
NCVVector<HaarFeature64> &h_HaarFeatures);
//! @}
#endif // _ncvhaarobjectdetection_hpp_
modules/cudalegacy/include/opencv2/cudalegacy/NCVPyramid.hpp
+113
View File
@@ -0,0 +1,113 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncvpyramid_hpp_
#define _ncvpyramid_hpp_
#include <memory>
#include <vector>
#include "opencv2/cudalegacy/NCV.hpp"
#include "opencv2/core/cuda/common.hpp"
//! @cond IGNORED
namespace cv { namespace cuda { namespace device
{
namespace pyramid
{
CV_EXPORTS void downsampleX2(PtrStepSzb src, PtrStepSzb dst, int depth, int cn, cudaStream_t stream);
CV_EXPORTS void interpolateFrom1(PtrStepSzb src, PtrStepSzb dst, int depth, int cn, cudaStream_t stream);
}
}}}
#if 0 //def _WIN32
template <class T>
class CV_EXPORTS NCVMatrixStack
{
public:
NCVMatrixStack() {this->_arr.clear();}
~NCVMatrixStack()
{
const Ncv32u nElem = this->_arr.size();
for (Ncv32u i=0; i<nElem; i++)
{
pop_back();
}
}
void push_back(NCVMatrix<T> *elem) {this->_arr.push_back(std::shared_ptr< NCVMatrix<T> >(elem));}
void pop_back() {this->_arr.pop_back();}
NCVMatrix<T> * operator [] (int i) const {return this->_arr[i].get();}
private:
std::vector< std::shared_ptr< NCVMatrix<T> > > _arr;
};
template <class T>
class CV_EXPORTS NCVImagePyramid
{
public:
NCVImagePyramid(const NCVMatrix<T> &img,
Ncv8u nLayers,
INCVMemAllocator &alloc,
cudaStream_t cuStream);
~NCVImagePyramid();
NcvBool isInitialized() const;
NCVStatus getLayer(NCVMatrix<T> &outImg,
NcvSize32u outRoi,
NcvBool bTrilinear,
cudaStream_t cuStream) const;
private:
NcvBool _isInitialized;
const NCVMatrix<T> *layer0;
NCVMatrixStack<T> pyramid;
Ncv32u nLayers;
};
#endif //_WIN32
//! @endcond
#endif //_ncvpyramid_hpp_
modules/cudalegacy/include/opencv2/cudalegacy/NPP_staging.hpp
+906
View File
@@ -0,0 +1,906 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _npp_staging_hpp_
#define _npp_staging_hpp_
#include "opencv2/cudalegacy/NCV.hpp"
//! @addtogroup cudalegacy
//! @{
/** \defgroup core_npp NPPST Core
* Basic functions for CUDA streams management.
* @{
*/
/**
* Gets an active CUDA stream used by NPPST
* NOT THREAD SAFE
* \return Current CUDA stream
*/
CV_EXPORTS
cudaStream_t nppStGetActiveCUDAstream();
/**
* Sets an active CUDA stream used by NPPST
* NOT THREAD SAFE
* \param cudaStream [IN] cudaStream CUDA stream to become current
* \return CUDA stream used before
*/
CV_EXPORTS
cudaStream_t nppStSetActiveCUDAstream(cudaStream_t cudaStream);
/** @} */
/** \defgroup nppi NPPST Image Processing
* @{
*/
/** Border type
*
* Filtering operations assume that each pixel has a neighborhood of pixels.
* The following structure describes possible ways to define non-existent pixels.
*/
enum NppStBorderType
{
nppStBorderNone = 0, ///< There is no need to define additional pixels, image is extended already
nppStBorderClamp = 1, ///< Clamp out of range position to borders
nppStBorderWrap = 2, ///< Wrap out of range position. Image becomes periodic.
nppStBorderMirror = 3 ///< reflect out of range position across borders
};
/**
* Filter types for image resizing
*/
enum NppStInterpMode
{
nppStSupersample, ///< Supersampling. For downscaling only
nppStBicubic ///< Bicubic convolution filter, a = -0.5 (cubic Hermite spline)
};
/** Frame interpolation state
*
* This structure holds parameters required for frame interpolation.
* Forward displacement field is a per-pixel mapping from frame 0 to frame 1.
* Backward displacement field is a per-pixel mapping from frame 1 to frame 0.
*/
struct NppStInterpolationState
{
NcvSize32u size; ///< frame size
Ncv32u nStep; ///< pitch
Ncv32f pos; ///< new frame position
Ncv32f *pSrcFrame0; ///< frame 0
Ncv32f *pSrcFrame1; ///< frame 1
Ncv32f *pFU; ///< forward horizontal displacement
Ncv32f *pFV; ///< forward vertical displacement
Ncv32f *pBU; ///< backward horizontal displacement
Ncv32f *pBV; ///< backward vertical displacement
Ncv32f *pNewFrame; ///< new frame
Ncv32f *ppBuffers[6]; ///< temporary buffers
};
/** Size of a buffer required for interpolation.
*
* Requires several such buffers. See \see NppStInterpolationState.
*
* \param srcSize [IN] Frame size (both frames must be of the same size)
* \param nStep [IN] Frame line step
* \param hpSize [OUT] Where to store computed size (host memory)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStGetInterpolationBufferSize(NcvSize32u srcSize,
Ncv32u nStep,
Ncv32u *hpSize);
/** Interpolate frames (images) using provided optical flow (displacement field).
* 32-bit floating point images, single channel
*
* \param pState [IN] structure containing all required parameters (host memory)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStInterpolateFrames(const NppStInterpolationState *pState);
/** Row linear filter. 32-bit floating point image, single channel
*
* Apply horizontal linear filter
*
* \param pSrc [IN] Source image pointer (CUDA device memory)
* \param srcSize [IN] Source image size
* \param nSrcStep [IN] Source image line step
* \param pDst [OUT] Destination image pointer (CUDA device memory)
* \param dstSize [OUT] Destination image size
* \param nDstStep
* \param oROI [IN] Region of interest in the source image
* \param borderType [IN] Type of border
* \param pKernel [IN] Pointer to row kernel values (CUDA device memory)
* \param nKernelSize [IN] Size of the kernel in pixels
* \param nAnchor [IN] The kernel row alignment with respect to the position of the input pixel
* \param multiplier [IN] Value by which the computed result is multiplied
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStFilterRowBorder_32f_C1R(Ncv32f *pSrc,
NcvSize32u srcSize,
Ncv32u nSrcStep,
Ncv32f *pDst,
NcvSize32u dstSize,
Ncv32u nDstStep,
NcvRect32u oROI,
NppStBorderType borderType,
Ncv32f *pKernel,
Ncv32s nKernelSize,
Ncv32s nAnchor,
Ncv32f multiplier);
/** Column linear filter. 32-bit floating point image, single channel
*
* Apply vertical linear filter
*
* \param pSrc [IN] Source image pointer (CUDA device memory)
* \param srcSize [IN] Source image size
* \param nSrcStep [IN] Source image line step
* \param pDst [OUT] Destination image pointer (CUDA device memory)
* \param dstSize [OUT] Destination image size
* \param nDstStep [IN]
* \param oROI [IN] Region of interest in the source image
* \param borderType [IN] Type of border
* \param pKernel [IN] Pointer to column kernel values (CUDA device memory)
* \param nKernelSize [IN] Size of the kernel in pixels
* \param nAnchor [IN] The kernel column alignment with respect to the position of the input pixel
* \param multiplier [IN] Value by which the computed result is multiplied
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStFilterColumnBorder_32f_C1R(Ncv32f *pSrc,
NcvSize32u srcSize,
Ncv32u nSrcStep,
Ncv32f *pDst,
NcvSize32u dstSize,
Ncv32u nDstStep,
NcvRect32u oROI,
NppStBorderType borderType,
Ncv32f *pKernel,
Ncv32s nKernelSize,
Ncv32s nAnchor,
Ncv32f multiplier);
/** Size of buffer required for vector image warping.
*
* \param srcSize [IN] Source image size
* \param nSrcStep [IN] Source image line step
* \param hpSize [OUT] Where to store computed size (host memory)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStVectorWarpGetBufferSize(NcvSize32u srcSize,
Ncv32u nSrcStep,
Ncv32u *hpSize);
/** Warp image using provided 2D vector field and 1x1 point spread function.
* 32-bit floating point image, single channel
*
* During warping pixels from the source image may fall between pixels of the destination image.
* PSF (point spread function) describes how the source image pixel affects pixels of the destination.
* For 1x1 PSF only single pixel with the largest intersection is affected (similar to nearest interpolation).
*
* Destination image size and line step must be the same as the source image size and line step
*
* \param pSrc [IN] Source image pointer (CUDA device memory)
* \param srcSize [IN] Source image size
* \param nSrcStep [IN] Source image line step
* \param pU [IN] Pointer to horizontal displacement field (CUDA device memory)
* \param pV [IN] Pointer to vertical displacement field (CUDA device memory)
* \param nVFStep [IN] Displacement field line step
* \param timeScale [IN] Value by which displacement field will be scaled for warping
* \param pDst [OUT] Destination image pointer (CUDA device memory)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStVectorWarp_PSF1x1_32f_C1(const Ncv32f *pSrc,
NcvSize32u srcSize,
Ncv32u nSrcStep,
const Ncv32f *pU,
const Ncv32f *pV,
Ncv32u nVFStep,
Ncv32f timeScale,
Ncv32f *pDst);
/** Warp image using provided 2D vector field and 2x2 point spread function.
* 32-bit floating point image, single channel
*
* During warping pixels from the source image may fall between pixels of the destination image.
* PSF (point spread function) describes how the source image pixel affects pixels of the destination.
* For 2x2 PSF all four intersected pixels will be affected.
*
* Destination image size and line step must be the same as the source image size and line step
*
* \param pSrc [IN] Source image pointer (CUDA device memory)
* \param srcSize [IN] Source image size
* \param nSrcStep [IN] Source image line step
* \param pU [IN] Pointer to horizontal displacement field (CUDA device memory)
* \param pV [IN] Pointer to vertical displacement field (CUDA device memory)
* \param nVFStep [IN] Displacement field line step
* \param pBuffer
* \param timeScale [IN] Value by which displacement field will be scaled for warping
* \param pDst [OUT] Destination image pointer (CUDA device memory)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStVectorWarp_PSF2x2_32f_C1(const Ncv32f *pSrc,
NcvSize32u srcSize,
Ncv32u nSrcStep,
const Ncv32f *pU,
const Ncv32f *pV,
Ncv32u nVFStep,
Ncv32f *pBuffer,
Ncv32f timeScale,
Ncv32f *pDst);
/** Resize. 32-bit floating point image, single channel
*
* Resizes image using specified filter (interpolation type)
*
* \param pSrc [IN] Source image pointer (CUDA device memory)
* \param srcSize [IN] Source image size
* \param nSrcStep [IN] Source image line step
* \param srcROI [IN] Source image region of interest
* \param pDst [OUT] Destination image pointer (CUDA device memory)
* \param dstSize [IN] Destination image size
* \param nDstStep [IN] Destination image line step
* \param dstROI [IN] Destination image region of interest
* \param xFactor [IN] Row scale factor
* \param yFactor [IN] Column scale factor
* \param interpolation [IN] Interpolation type
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStResize_32f_C1R(Ncv32f *pSrc,
NcvSize32u srcSize,
Ncv32u nSrcStep,
NcvRect32u srcROI,
Ncv32f *pDst,
NcvSize32u dstSize,
Ncv32u nDstStep,
NcvRect32u dstROI,
Ncv32f xFactor,
Ncv32f yFactor,
NppStInterpMode interpolation);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit unsigned pixels, single channel.
*
* \param d_src [IN] Source image pointer (CUDA device memory)
* \param srcStep [IN] Source image line step
* \param d_dst [OUT] Destination image pointer (CUDA device memory)
* \param dstStep [IN] Destination image line step
* \param srcRoi [IN] Region of interest in the source image
* \param scale [IN] Downsampling scale factor (positive integer)
* \param readThruTexture [IN] Performance hint to cache source in texture (true) or read directly (false)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStDecimate_32u_C1R(Ncv32u *d_src, Ncv32u srcStep,
Ncv32u *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit signed pixels, single channel.
* \see nppiStDecimate_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStDecimate_32s_C1R(Ncv32s *d_src, Ncv32u srcStep,
Ncv32s *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit float pixels, single channel.
* \see nppiStDecimate_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStDecimate_32f_C1R(Ncv32f *d_src, Ncv32u srcStep,
Ncv32f *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit unsigned pixels, single channel.
* \see nppiStDecimate_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStDecimate_64u_C1R(Ncv64u *d_src, Ncv32u srcStep,
Ncv64u *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit signed pixels, single channel.
* \see nppiStDecimate_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStDecimate_64s_C1R(Ncv64s *d_src, Ncv32u srcStep,
Ncv64s *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit float pixels, single channel.
* \see nppiStDecimate_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStDecimate_64f_C1R(Ncv64f *d_src, Ncv32u srcStep,
Ncv64f *d_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale,
NcvBool readThruTexture);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit unsigned pixels, single channel. Host implementation.
*
* \param h_src [IN] Source image pointer (Host or pinned memory)
* \param srcStep [IN] Source image line step
* \param h_dst [OUT] Destination image pointer (Host or pinned memory)
* \param dstStep [IN] Destination image line step
* \param srcRoi [IN] Region of interest in the source image
* \param scale [IN] Downsampling scale factor (positive integer)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStDecimate_32u_C1R_host(Ncv32u *h_src, Ncv32u srcStep,
Ncv32u *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit signed pixels, single channel. Host implementation.
* \see nppiStDecimate_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStDecimate_32s_C1R_host(Ncv32s *h_src, Ncv32u srcStep,
Ncv32s *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 32-bit float pixels, single channel. Host implementation.
* \see nppiStDecimate_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStDecimate_32f_C1R_host(Ncv32f *h_src, Ncv32u srcStep,
Ncv32f *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit unsigned pixels, single channel. Host implementation.
* \see nppiStDecimate_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStDecimate_64u_C1R_host(Ncv64u *h_src, Ncv32u srcStep,
Ncv64u *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit signed pixels, single channel. Host implementation.
* \see nppiStDecimate_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStDecimate_64s_C1R_host(Ncv64s *h_src, Ncv32u srcStep,
Ncv64s *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Downsamples (decimates) an image using the nearest neighbor algorithm. 64-bit float pixels, single channel. Host implementation.
* \see nppiStDecimate_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStDecimate_64f_C1R_host(Ncv64f *h_src, Ncv32u srcStep,
Ncv64f *h_dst, Ncv32u dstStep,
NcvSize32u srcRoi, Ncv32u scale);
/**
* Computes standard deviation for each rectangular region of the input image using integral images.
*
* \param d_sum [IN] Integral image pointer (CUDA device memory)
* \param sumStep [IN] Integral image line step
* \param d_sqsum [IN] Squared integral image pointer (CUDA device memory)
* \param sqsumStep [IN] Squared integral image line step
* \param d_norm [OUT] Stddev image pointer (CUDA device memory). Each pixel contains stddev of a rect with top-left corner at the original location in the image
* \param normStep [IN] Stddev image line step
* \param roi [IN] Region of interest in the source image
* \param rect [IN] Rectangular region to calculate stddev over
* \param scaleArea [IN] Multiplication factor to account decimated scale
* \param readThruTexture [IN] Performance hint to cache source in texture (true) or read directly (false)
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStRectStdDev_32f_C1R(Ncv32u *d_sum, Ncv32u sumStep,
Ncv64u *d_sqsum, Ncv32u sqsumStep,
Ncv32f *d_norm, Ncv32u normStep,
NcvSize32u roi, NcvRect32u rect,
Ncv32f scaleArea, NcvBool readThruTexture);
/**
* Computes standard deviation for each rectangular region of the input image using integral images. Host implementation
*
* \param h_sum [IN] Integral image pointer (Host or pinned memory)
* \param sumStep [IN] Integral image line step
* \param h_sqsum [IN] Squared integral image pointer (Host or pinned memory)
* \param sqsumStep [IN] Squared integral image line step
* \param h_norm [OUT] Stddev image pointer (Host or pinned memory). Each pixel contains stddev of a rect with top-left corner at the original location in the image
* \param normStep [IN] Stddev image line step
* \param roi [IN] Region of interest in the source image
* \param rect [IN] Rectangular region to calculate stddev over
* \param scaleArea [IN] Multiplication factor to account decimated scale
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStRectStdDev_32f_C1R_host(Ncv32u *h_sum, Ncv32u sumStep,
Ncv64u *h_sqsum, Ncv32u sqsumStep,
Ncv32f *h_norm, Ncv32u normStep,
NcvSize32u roi, NcvRect32u rect,
Ncv32f scaleArea);
/**
* Transposes an image. 32-bit unsigned pixels, single channel
*
* \param d_src [IN] Source image pointer (CUDA device memory)
* \param srcStride [IN] Source image line step
* \param d_dst [OUT] Destination image pointer (CUDA device memory)
* \param dstStride [IN] Destination image line step
* \param srcRoi [IN] Region of interest of the source image
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStTranspose_32u_C1R(Ncv32u *d_src, Ncv32u srcStride,
Ncv32u *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 32-bit signed pixels, single channel
* \see nppiStTranspose_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStTranspose_32s_C1R(Ncv32s *d_src, Ncv32u srcStride,
Ncv32s *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 32-bit float pixels, single channel
* \see nppiStTranspose_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStTranspose_32f_C1R(Ncv32f *d_src, Ncv32u srcStride,
Ncv32f *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 64-bit unsigned pixels, single channel
* \see nppiStTranspose_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStTranspose_64u_C1R(Ncv64u *d_src, Ncv32u srcStride,
Ncv64u *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 64-bit signed pixels, single channel
* \see nppiStTranspose_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStTranspose_64s_C1R(Ncv64s *d_src, Ncv32u srcStride,
Ncv64s *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 64-bit float pixels, single channel
* \see nppiStTranspose_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStTranspose_64f_C1R(Ncv64f *d_src, Ncv32u srcStride,
Ncv64f *d_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 128-bit pixels of any type, single channel
* \see nppiStTranspose_32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStTranspose_128_C1R(void *d_src, Ncv32u srcStep,
void *d_dst, Ncv32u dstStep, NcvSize32u srcRoi);
/**
* Transposes an image. 32-bit unsigned pixels, single channel. Host implementation
*
* \param h_src [IN] Source image pointer (Host or pinned memory)
* \param srcStride [IN] Source image line step
* \param h_dst [OUT] Destination image pointer (Host or pinned memory)
* \param dstStride [IN] Destination image line step
* \param srcRoi [IN] Region of interest of the source image
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStTranspose_32u_C1R_host(Ncv32u *h_src, Ncv32u srcStride,
Ncv32u *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 32-bit signed pixels, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStTranspose_32s_C1R_host(Ncv32s *h_src, Ncv32u srcStride,
Ncv32s *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 32-bit float pixels, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStTranspose_32f_C1R_host(Ncv32f *h_src, Ncv32u srcStride,
Ncv32f *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 64-bit unsigned pixels, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStTranspose_64u_C1R_host(Ncv64u *h_src, Ncv32u srcStride,
Ncv64u *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 64-bit signed pixels, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStTranspose_64s_C1R_host(Ncv64s *h_src, Ncv32u srcStride,
Ncv64s *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 64-bit float pixels, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStTranspose_64f_C1R_host(Ncv64f *h_src, Ncv32u srcStride,
Ncv64f *h_dst, Ncv32u dstStride, NcvSize32u srcRoi);
/**
* Transposes an image. 128-bit pixels of any type, single channel. Host implementation
* \see nppiStTranspose_32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStTranspose_128_C1R_host(void *d_src, Ncv32u srcStep,
void *d_dst, Ncv32u dstStep, NcvSize32u srcRoi);
/**
* Calculates the size of the temporary buffer for integral image creation
*
* \param roiSize [IN] Size of the input image
* \param pBufsize [OUT] Pointer to host variable that returns the size of the temporary buffer (in bytes)
* \param devProp [IN] CUDA device properties structure, containing texture alignment information
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStIntegralGetSize_8u32u(NcvSize32u roiSize, Ncv32u *pBufsize, cudaDeviceProp &devProp);
/**
* Calculates the size of the temporary buffer for integral image creation
* \see nppiStIntegralGetSize_8u32u
*/
CV_EXPORTS
NCVStatus nppiStIntegralGetSize_32f32f(NcvSize32u roiSize, Ncv32u *pBufsize, cudaDeviceProp &devProp);
/**
* Creates an integral image representation for the input image
*
* \param d_src [IN] Source image pointer (CUDA device memory)
* \param srcStep [IN] Source image line step
* \param d_dst [OUT] Destination integral image pointer (CUDA device memory)
* \param dstStep [IN] Destination image line step
* \param roiSize [IN] Region of interest of the source image
* \param pBuffer [IN] Pointer to the pre-allocated temporary buffer (CUDA device memory)
* \param bufSize [IN] Size of the pBuffer in bytes
* \param devProp [IN] CUDA device properties structure, containing texture alignment information
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStIntegral_8u32u_C1R(Ncv8u *d_src, Ncv32u srcStep,
Ncv32u *d_dst, Ncv32u dstStep, NcvSize32u roiSize,
Ncv8u *pBuffer, Ncv32u bufSize, cudaDeviceProp &devProp);
/**
* Creates an integral image representation for the input image
* \see nppiStIntegral_8u32u_C1R
*/
CV_EXPORTS
NCVStatus nppiStIntegral_32f32f_C1R(Ncv32f *d_src, Ncv32u srcStep,
Ncv32f *d_dst, Ncv32u dstStep, NcvSize32u roiSize,
Ncv8u *pBuffer, Ncv32u bufSize, cudaDeviceProp &devProp);
/**
* Creates an integral image representation for the input image. Host implementation
*
* \param h_src [IN] Source image pointer (Host or pinned memory)
* \param srcStep [IN] Source image line step
* \param h_dst [OUT] Destination integral image pointer (Host or pinned memory)
* \param dstStep [IN] Destination image line step
* \param roiSize [IN] Region of interest of the source image
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStIntegral_8u32u_C1R_host(Ncv8u *h_src, Ncv32u srcStep,
Ncv32u *h_dst, Ncv32u dstStep, NcvSize32u roiSize);
/**
* Creates an integral image representation for the input image. Host implementation
* \see nppiStIntegral_8u32u_C1R_host
*/
CV_EXPORTS
NCVStatus nppiStIntegral_32f32f_C1R_host(Ncv32f *h_src, Ncv32u srcStep,
Ncv32f *h_dst, Ncv32u dstStep, NcvSize32u roiSize);
/**
* Calculates the size of the temporary buffer for squared integral image creation
*
* \param roiSize [IN] Size of the input image
* \param pBufsize [OUT] Pointer to host variable that returns the size of the temporary buffer (in bytes)
* \param devProp [IN] CUDA device properties structure, containing texture alignment information
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStSqrIntegralGetSize_8u64u(NcvSize32u roiSize, Ncv32u *pBufsize, cudaDeviceProp &devProp);
/**
* Creates a squared integral image representation for the input image
*
* \param d_src [IN] Source image pointer (CUDA device memory)
* \param srcStep [IN] Source image line step
* \param d_dst [OUT] Destination squared integral image pointer (CUDA device memory)
* \param dstStep [IN] Destination image line step
* \param roiSize [IN] Region of interest of the source image
* \param pBuffer [IN] Pointer to the pre-allocated temporary buffer (CUDA device memory)
* \param bufSize [IN] Size of the pBuffer in bytes
* \param devProp [IN] CUDA device properties structure, containing texture alignment information
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStSqrIntegral_8u64u_C1R(Ncv8u *d_src, Ncv32u srcStep,
Ncv64u *d_dst, Ncv32u dstStep, NcvSize32u roiSize,
Ncv8u *pBuffer, Ncv32u bufSize, cudaDeviceProp &devProp);
/**
* Creates a squared integral image representation for the input image. Host implementation
*
* \param h_src [IN] Source image pointer (Host or pinned memory)
* \param srcStep [IN] Source image line step
* \param h_dst [OUT] Destination squared integral image pointer (Host or pinned memory)
* \param dstStep [IN] Destination image line step
* \param roiSize [IN] Region of interest of the source image
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppiStSqrIntegral_8u64u_C1R_host(Ncv8u *h_src, Ncv32u srcStep,
Ncv64u *h_dst, Ncv32u dstStep, NcvSize32u roiSize);
/** @} */
/** \defgroup npps NPPST Signal Processing
* @{
*/
/**
* Calculates the size of the temporary buffer for vector compaction. 32-bit unsigned values
*
* \param srcLen [IN] Length of the input vector in elements
* \param pBufsize [OUT] Pointer to host variable that returns the size of the temporary buffer (in bytes)
* \param devProp [IN] CUDA device properties structure, containing texture alignment information
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppsStCompactGetSize_32u(Ncv32u srcLen, Ncv32u *pBufsize, cudaDeviceProp &devProp);
/**
* Calculates the size of the temporary buffer for vector compaction. 32-bit signed values
* \see nppsStCompactGetSize_32u
*/
NCVStatus nppsStCompactGetSize_32s(Ncv32u srcLen, Ncv32u *pBufsize, cudaDeviceProp &devProp);
/**
* Calculates the size of the temporary buffer for vector compaction. 32-bit float values
* \see nppsStCompactGetSize_32u
*/
NCVStatus nppsStCompactGetSize_32f(Ncv32u srcLen, Ncv32u *pBufsize, cudaDeviceProp &devProp);
/**
* Compacts the input vector by removing elements of specified value. 32-bit unsigned values
*
* \param d_src [IN] Source vector pointer (CUDA device memory)
* \param srcLen [IN] Source vector length
* \param d_dst [OUT] Destination vector pointer (CUDA device memory)
* \param p_dstLen [OUT] Pointer to the destination vector length (Pinned memory or NULL)
* \param elemRemove [IN] The value to be removed
* \param pBuffer [IN] Pointer to the pre-allocated temporary buffer (CUDA device memory)
* \param bufSize [IN] Size of the pBuffer in bytes
* \param devProp [IN] CUDA device properties structure, containing texture alignment information
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppsStCompact_32u(Ncv32u *d_src, Ncv32u srcLen,
Ncv32u *d_dst, Ncv32u *p_dstLen,
Ncv32u elemRemove, Ncv8u *pBuffer,
Ncv32u bufSize, cudaDeviceProp &devProp);
/**
* Compacts the input vector by removing elements of specified value. 32-bit signed values
* \see nppsStCompact_32u
*/
CV_EXPORTS
NCVStatus nppsStCompact_32s(Ncv32s *d_src, Ncv32u srcLen,
Ncv32s *d_dst, Ncv32u *p_dstLen,
Ncv32s elemRemove, Ncv8u *pBuffer,
Ncv32u bufSize, cudaDeviceProp &devProp);
/**
* Compacts the input vector by removing elements of specified value. 32-bit float values
* \see nppsStCompact_32u
*/
CV_EXPORTS
NCVStatus nppsStCompact_32f(Ncv32f *d_src, Ncv32u srcLen,
Ncv32f *d_dst, Ncv32u *p_dstLen,
Ncv32f elemRemove, Ncv8u *pBuffer,
Ncv32u bufSize, cudaDeviceProp &devProp);
/**
* Compacts the input vector by removing elements of specified value. 32-bit unsigned values. Host implementation
*
* \param h_src [IN] Source vector pointer (CUDA device memory)
* \param srcLen [IN] Source vector length
* \param h_dst [OUT] Destination vector pointer (CUDA device memory)
* \param dstLen [OUT] Pointer to the destination vector length (can be NULL)
* \param elemRemove [IN] The value to be removed
*
* \return NCV status code
*/
CV_EXPORTS
NCVStatus nppsStCompact_32u_host(Ncv32u *h_src, Ncv32u srcLen,
Ncv32u *h_dst, Ncv32u *dstLen, Ncv32u elemRemove);
/**
* Compacts the input vector by removing elements of specified value. 32-bit signed values. Host implementation
* \see nppsStCompact_32u_host
*/
CV_EXPORTS
NCVStatus nppsStCompact_32s_host(Ncv32s *h_src, Ncv32u srcLen,
Ncv32s *h_dst, Ncv32u *dstLen, Ncv32s elemRemove);
/**
* Compacts the input vector by removing elements of specified value. 32-bit float values. Host implementation
* \see nppsStCompact_32u_host
*/
CV_EXPORTS
NCVStatus nppsStCompact_32f_host(Ncv32f *h_src, Ncv32u srcLen,
Ncv32f *h_dst, Ncv32u *dstLen, Ncv32f elemRemove);
/** @} */
//! @} cudalegacy
#endif // _npp_staging_hpp_
modules/cudalegacy/include/opencv2/cudalegacy/private.hpp
+96
View File
@@ -0,0 +1,96 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef OPENCV_CORE_CUDALEGACY_PRIVATE_HPP
#define OPENCV_CORE_CUDALEGACY_PRIVATE_HPP
#ifndef __OPENCV_BUILD
# error this is a private header which should not be used from outside of the OpenCV library
#endif
#include "opencv2/core/private.cuda.hpp"
#ifdef HAVE_CUDA
#include "opencv2/cudalegacy.hpp"
//! @cond IGNORED
namespace cv { namespace cuda
{
class NppStStreamHandler
{
public:
inline explicit NppStStreamHandler(cudaStream_t newStream = 0)
{
oldStream = nppStSetActiveCUDAstream(newStream);
}
inline ~NppStStreamHandler()
{
nppStSetActiveCUDAstream(oldStream);
}
private:
cudaStream_t oldStream;
};
CV_EXPORTS cv::String getNcvErrorMessage(int code);
static inline void checkNcvError(int err, const char* file, const int line, const char* func)
{
if (NCV_SUCCESS != err)
{
cv::String msg = getNcvErrorMessage(err);
cv::error(cv::Error::GpuApiCallError, msg, func, file, line);
}
}
}}
#define ncvSafeCall(expr) cv::cuda::checkNcvError(expr, __FILE__, __LINE__, CV_Func)
//! @endcond
#endif // HAVE_CUDA
#endif // OPENCV_CORE_CUDALEGACY_PRIVATE_HPP
modules/cudalegacy/perf/perf_bgsegm.cpp
+236
View File
@@ -0,0 +1,236 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
#ifdef HAVE_OPENCV_CUDAIMGPROC
# include "opencv2/cudaimgproc.hpp"
#endif
namespace opencv_test { namespace {
//////////////////////////////////////////////////////
// FGDStatModel
#ifdef HAVE_VIDEO_INPUT
DEF_PARAM_TEST_1(Video, string);
PERF_TEST_P(Video, FGDStatModel,
Values(string("cv/video/768x576.avi")))
{
const int numIters = 10;
declare.time(60);
const string inputFile = perf::TestBase::getDataPath(GetParam());
cv::VideoCapture cap(inputFile);
ASSERT_TRUE(cap.isOpened());
cv::Mat frame;
cap >> frame;
ASSERT_FALSE(frame.empty());
if (PERF_RUN_CUDA())
{
cv::cuda::GpuMat d_frame(frame), foreground;
cv::Ptr<cv::cuda::BackgroundSubtractorFGD> d_fgd = cv::cuda::createBackgroundSubtractorFGD();
d_fgd->apply(d_frame, foreground);
int i = 0;
// collect performance data
for (; i < numIters; ++i)
{
cap >> frame;
ASSERT_FALSE(frame.empty());
d_frame.upload(frame);
startTimer();
if(!next())
break;
d_fgd->apply(d_frame, foreground);
stopTimer();
}
// process last frame in sequence to get data for sanity test
for (; i < numIters; ++i)
{
cap >> frame;
ASSERT_FALSE(frame.empty());
d_frame.upload(frame);
d_fgd->apply(d_frame, foreground);
}
}
else
{
FAIL_NO_CPU();
}
SANITY_CHECK_NOTHING();
}
#endif
//////////////////////////////////////////////////////
// GMG
#ifdef HAVE_VIDEO_INPUT
DEF_PARAM_TEST(Video_Cn_MaxFeatures, string, MatCn, int);
PERF_TEST_P(Video_Cn_MaxFeatures, GMG,
Combine(Values(string("cv/video/768x576.avi")),
CUDA_CHANNELS_1_3_4,
Values(20, 40, 60)))
{
const int numIters = 150;
const std::string inputFile = perf::TestBase::getDataPath(GET_PARAM(0));
const int cn = GET_PARAM(1);
const int maxFeatures = GET_PARAM(2);
cv::VideoCapture cap(inputFile);
ASSERT_TRUE(cap.isOpened());
cv::Mat frame;
cap >> frame;
ASSERT_FALSE(frame.empty());
if (cn != 3)
{
cv::Mat temp;
if (cn == 1)
cv::cvtColor(frame, temp, cv::COLOR_BGR2GRAY);
else
cv::cvtColor(frame, temp, cv::COLOR_BGR2BGRA);
cv::swap(temp, frame);
}
if (PERF_RUN_CUDA())
{
cv::cuda::GpuMat d_frame(frame);
cv::cuda::GpuMat foreground;
cv::Ptr<cv::cuda::BackgroundSubtractorGMG> d_gmg = cv::cuda::createBackgroundSubtractorGMG();
d_gmg->setMaxFeatures(maxFeatures);
d_gmg->apply(d_frame, foreground);
int i = 0;
// collect performance data
for (; i < numIters; ++i)
{
cap >> frame;
if (frame.empty())
{
cap.release();
cap.open(inputFile);
cap >> frame;
}
if (cn != 3)
{
cv::Mat temp;
if (cn == 1)
cv::cvtColor(frame, temp, cv::COLOR_BGR2GRAY);
else
cv::cvtColor(frame, temp, cv::COLOR_BGR2BGRA);
cv::swap(temp, frame);
}
d_frame.upload(frame);
startTimer();
if(!next())
break;
d_gmg->apply(d_frame, foreground);
stopTimer();
}
// process last frame in sequence to get data for sanity test
for (; i < numIters; ++i)
{
cap >> frame;
if (frame.empty())
{
cap.release();
cap.open(inputFile);
cap >> frame;
}
if (cn != 3)
{
cv::Mat temp;
if (cn == 1)
cv::cvtColor(frame, temp, cv::COLOR_BGR2GRAY);
else
cv::cvtColor(frame, temp, cv::COLOR_BGR2BGRA);
cv::swap(temp, frame);
}
d_frame.upload(frame);
d_gmg->apply(d_frame, foreground);
}
}
else
{
FAIL_NO_CPU();
}
SANITY_CHECK_NOTHING();
}
#endif
}} // namespace
modules/cudalegacy/perf/perf_calib3d.cpp
+140
View File
@@ -0,0 +1,140 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
#ifdef HAVE_OPENCV_3D
#include "opencv2/geometry.hpp"
namespace opencv_test { namespace {
DEF_PARAM_TEST_1(Count, int);
//////////////////////////////////////////////////////////////////////
// ProjectPoints
PERF_TEST_P(Count, Calib3D_ProjectPoints,
Values(5000, 10000, 20000))
{
const int count = GetParam();
cv::Mat src(1, count, CV_32FC3);
declare.in(src, WARMUP_RNG);
const cv::Mat rvec = cv::Mat::ones(1, 3, CV_32FC1);
const cv::Mat tvec = cv::Mat::ones(1, 3, CV_32FC1);
const cv::Mat camera_mat = cv::Mat::ones(3, 3, CV_32FC1);
if (PERF_RUN_CUDA())
{
const cv::cuda::GpuMat d_src(src);
cv::cuda::GpuMat dst;
TEST_CYCLE() cv::cuda::projectPoints(d_src, rvec, tvec, camera_mat, cv::Mat(), dst);
CUDA_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
TEST_CYCLE() cv::projectPoints(src, rvec, tvec, camera_mat, cv::noArray(), dst);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// SolvePnPRansac
PERF_TEST_P(Count, Calib3D_SolvePnPRansac,
Values(5000, 10000, 20000))
{
declare.time(10.0);
const int count = GetParam();
cv::Mat object(1, count, CV_32FC3);
declare.in(object, WARMUP_RNG);
cv::Mat camera_mat(3, 3, CV_32FC1);
cv::randu(camera_mat, 0.5, 1);
camera_mat.at<float>(0, 1) = 0.f;
camera_mat.at<float>(1, 0) = 0.f;
camera_mat.at<float>(2, 0) = 0.f;
camera_mat.at<float>(2, 1) = 0.f;
const cv::Mat dist_coef(1, 8, CV_32F, cv::Scalar::all(0));
cv::Mat rvec_gold(1, 3, CV_32FC1);
cv::randu(rvec_gold, 0, 1);
cv::Mat tvec_gold(1, 3, CV_32FC1);
cv::randu(tvec_gold, 0, 1);
std::vector<cv::Point2f> image_vec;
cv::projectPoints(object, rvec_gold, tvec_gold, camera_mat, dist_coef, image_vec);
const cv::Mat image(1, count, CV_32FC2, &image_vec[0]);
cv::Mat rvec;
cv::Mat tvec;
if (PERF_RUN_CUDA())
{
TEST_CYCLE() cv::cuda::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
CUDA_SANITY_CHECK(rvec, 1e-3);
CUDA_SANITY_CHECK(tvec, 1e-3);
}
else
{
TEST_CYCLE() cv::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
CPU_SANITY_CHECK(rvec, 1e-6);
CPU_SANITY_CHECK(tvec, 1e-6);
}
}
}} // namespace
#endif
modules/cudalegacy/perf/perf_labeling.cpp
+195
View File
@@ -0,0 +1,195 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
namespace opencv_test { namespace {
DEF_PARAM_TEST_1(Image, string);
struct GreedyLabeling
{
struct dot
{
int x;
int y;
static dot make(int i, int j)
{
dot d; d.x = i; d.y = j;
return d;
}
};
struct InInterval
{
InInterval(const int& _lo, const int& _hi) : lo(-_lo), hi(_hi) {}
const int lo, hi;
bool operator() (const unsigned char a, const unsigned char b) const
{
int d = a - b;
return lo <= d && d <= hi;
}
private:
InInterval& operator=(const InInterval&);
};
GreedyLabeling(cv::Mat img)
: image(img), _labels(image.size(), CV_32SC1, cv::Scalar::all(-1)) {stack = new dot[image.cols * image.rows];}
~GreedyLabeling(){delete[] stack;}
void operator() (cv::Mat labels) const
{
labels.setTo(cv::Scalar::all(-1));
InInterval inInt(0, 2);
int cc = -1;
int* dist_labels = (int*)labels.data;
int pitch = static_cast<int>(labels.step1());
unsigned char* source = (unsigned char*)image.data;
int width = image.cols;
int height = image.rows;
for (int j = 0; j < image.rows; ++j)
for (int i = 0; i < image.cols; ++i)
{
if (dist_labels[j * pitch + i] != -1) continue;
dot* top = stack;
dot p = dot::make(i, j);
cc++;
dist_labels[j * pitch + i] = cc;
while (top >= stack)
{
int* dl = &dist_labels[p.y * pitch + p.x];
unsigned char* sp = &source[p.y * image.step1() + p.x];
dl[0] = cc;
//right
if( p.x < (width - 1) && dl[ +1] == -1 && inInt(sp[0], sp[+1]))
*top++ = dot::make(p.x + 1, p.y);
//left
if( p.x > 0 && dl[-1] == -1 && inInt(sp[0], sp[-1]))
*top++ = dot::make(p.x - 1, p.y);
//bottom
if( p.y < (height - 1) && dl[+pitch] == -1 && inInt(sp[0], sp[+image.step1()]))
*top++ = dot::make(p.x, p.y + 1);
//top
if( p.y > 0 && dl[-pitch] == -1 && inInt(sp[0], sp[-static_cast<int>(image.step1())]))
*top++ = dot::make(p.x, p.y - 1);
p = *--top;
}
}
}
cv::Mat image;
cv::Mat _labels;
dot* stack;
};
PERF_TEST_P(Image, DISABLED_Labeling_ConnectivityMask,
Values<string>("gpu/labeling/aloe-disp.png"))
{
declare.time(1.0);
const cv::Mat image = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
if (PERF_RUN_CUDA())
{
cv::cuda::GpuMat d_image(image);
cv::cuda::GpuMat mask;
TEST_CYCLE() cv::cuda::connectivityMask(d_image, mask, cv::Scalar::all(0), cv::Scalar::all(2));
CUDA_SANITY_CHECK(mask);
}
else
{
FAIL_NO_CPU();
}
}
PERF_TEST_P(Image, DISABLED_Labeling_ConnectedComponents,
Values<string>("gpu/labeling/aloe-disp.png"))
{
declare.time(1.0);
const cv::Mat image = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
if (PERF_RUN_CUDA())
{
cv::cuda::GpuMat d_mask;
cv::cuda::connectivityMask(cv::cuda::GpuMat(image), d_mask, cv::Scalar::all(0), cv::Scalar::all(2));
cv::cuda::GpuMat components;
TEST_CYCLE() cv::cuda::labelComponents(d_mask, components);
CUDA_SANITY_CHECK(components);
}
else
{
GreedyLabeling host(image);
TEST_CYCLE() host(host._labels);
cv::Mat components = host._labels;
CPU_SANITY_CHECK(components);
}
}
}} // namespace
modules/cudalegacy/perf/perf_main.cpp
+47
View File
@@ -0,0 +1,47 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
using namespace perf;
CV_PERF_TEST_CUDA_MAIN(cudalegacy)
modules/cudalegacy/perf/perf_precomp.hpp
+57
View File
@@ -0,0 +1,57 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_PERF_PRECOMP_HPP__
#define __OPENCV_PERF_PRECOMP_HPP__
#include "opencv2/ts.hpp"
#include "opencv2/ts/cuda_perf.hpp"
#include "opencv2/cudalegacy.hpp"
#include "opencv2/video.hpp"
#include "opencv2/opencv_modules.hpp"
namespace opencv_test {
using namespace perf;
}
#endif
modules/cudalegacy/src/NCV.cpp
+891
View File
@@ -0,0 +1,891 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#if defined (HAVE_CUDA) && !defined (CUDA_DISABLER)
//==============================================================================
//
// Error handling helpers
//
//==============================================================================
namespace
{
#define error_entry(entry) { entry, #entry }
struct ErrorEntry
{
int code;
const char* str;
};
struct ErrorEntryComparer
{
int code;
ErrorEntryComparer(int code_) : code(code_) {}
bool operator()(const ErrorEntry& e) const { return e.code == code; }
};
//////////////////////////////////////////////////////////////////////////
// NCV errors
const ErrorEntry ncv_errors [] =
{
error_entry( NCV_SUCCESS ),
error_entry( NCV_UNKNOWN_ERROR ),
error_entry( NCV_CUDA_ERROR ),
error_entry( NCV_NPP_ERROR ),
error_entry( NCV_FILE_ERROR ),
error_entry( NCV_NULL_PTR ),
error_entry( NCV_INCONSISTENT_INPUT ),
error_entry( NCV_TEXTURE_BIND_ERROR ),
error_entry( NCV_DIMENSIONS_INVALID ),
error_entry( NCV_INVALID_ROI ),
error_entry( NCV_INVALID_STEP ),
error_entry( NCV_INVALID_SCALE ),
error_entry( NCV_INVALID_SCALE ),
error_entry( NCV_ALLOCATOR_NOT_INITIALIZED ),
error_entry( NCV_ALLOCATOR_BAD_ALLOC ),
error_entry( NCV_ALLOCATOR_BAD_DEALLOC ),
error_entry( NCV_ALLOCATOR_INSUFFICIENT_CAPACITY ),
error_entry( NCV_ALLOCATOR_DEALLOC_ORDER ),
error_entry( NCV_ALLOCATOR_BAD_REUSE ),
error_entry( NCV_MEM_COPY_ERROR ),
error_entry( NCV_MEM_RESIDENCE_ERROR ),
error_entry( NCV_MEM_INSUFFICIENT_CAPACITY ),
error_entry( NCV_HAAR_INVALID_PIXEL_STEP ),
error_entry( NCV_HAAR_TOO_MANY_FEATURES_IN_CLASSIFIER ),
error_entry( NCV_HAAR_TOO_MANY_FEATURES_IN_CASCADE ),
error_entry( NCV_HAAR_TOO_LARGE_FEATURES ),
error_entry( NCV_HAAR_XML_LOADING_EXCEPTION ),
error_entry( NCV_NOIMPL_HAAR_TILTED_FEATURES ),
error_entry( NCV_WARNING_HAAR_DETECTIONS_VECTOR_OVERFLOW ),
error_entry( NPPST_SUCCESS ),
error_entry( NPPST_ERROR ),
error_entry( NPPST_CUDA_KERNEL_EXECUTION_ERROR ),
error_entry( NPPST_NULL_POINTER_ERROR ),
error_entry( NPPST_TEXTURE_BIND_ERROR ),
error_entry( NPPST_MEMCPY_ERROR ),
error_entry( NPPST_MEM_ALLOC_ERR ),
error_entry( NPPST_MEMFREE_ERR ),
error_entry( NPPST_INVALID_ROI ),
error_entry( NPPST_INVALID_STEP ),
error_entry( NPPST_INVALID_SCALE ),
error_entry( NPPST_MEM_INSUFFICIENT_BUFFER ),
error_entry( NPPST_MEM_RESIDENCE_ERROR ),
error_entry( NPPST_MEM_INTERNAL_ERROR )
};
const size_t ncv_error_num = sizeof(ncv_errors) / sizeof(ncv_errors[0]);
}
cv::String cv::cuda::getNcvErrorMessage(int code)
{
size_t idx = std::find_if(ncv_errors, ncv_errors + ncv_error_num, ErrorEntryComparer(code)) - ncv_errors;
const char* msg = (idx != ncv_error_num) ? ncv_errors[idx].str : "Unknown error code";
String str = cv::format("%s [Code = %d]", msg, code);
return str;
}
static void stdDebugOutput(const cv::String &msg)
{
std::cout << msg.c_str() << std::endl;
}
static NCVDebugOutputHandler *debugOutputHandler = stdDebugOutput;
void ncvDebugOutput(const cv::String &msg)
{
debugOutputHandler(msg);
}
void ncvSetDebugOutputHandler(NCVDebugOutputHandler *func)
{
debugOutputHandler = func;
}
//==============================================================================
//
// Memory wrappers and helpers
//
//==============================================================================
Ncv32u alignUp(Ncv32u what, Ncv32u alignment)
{
Ncv32u alignMask = alignment-1;
Ncv32u inverseAlignMask = ~alignMask;
Ncv32u res = (what + alignMask) & inverseAlignMask;
return res;
}
void NCVMemPtr::clear()
{
ptr = NULL;
memtype = NCVMemoryTypeNone;
}
void NCVMemSegment::clear()
{
begin.clear();
size = 0;
}
NCVStatus memSegCopyHelper(void *dst, NCVMemoryType dstType, const void *src, NCVMemoryType srcType, size_t sz, cudaStream_t cuStream)
{
NCVStatus ncvStat;
switch (dstType)
{
case NCVMemoryTypeHostPageable:
case NCVMemoryTypeHostPinned:
switch (srcType)
{
case NCVMemoryTypeHostPageable:
case NCVMemoryTypeHostPinned:
memcpy(dst, src, sz);
ncvStat = NCV_SUCCESS;
break;
case NCVMemoryTypeDevice:
if (cuStream != 0)
{
ncvAssertCUDAReturn(cudaMemcpyAsync(dst, src, sz, cudaMemcpyDeviceToHost, cuStream), NCV_CUDA_ERROR);
}
else
{
ncvAssertCUDAReturn(cudaMemcpy(dst, src, sz, cudaMemcpyDeviceToHost), NCV_CUDA_ERROR);
}
ncvStat = NCV_SUCCESS;
break;
default:
ncvStat = NCV_MEM_RESIDENCE_ERROR;
}
break;
case NCVMemoryTypeDevice:
switch (srcType)
{
case NCVMemoryTypeHostPageable:
case NCVMemoryTypeHostPinned:
if (cuStream != 0)
{
ncvAssertCUDAReturn(cudaMemcpyAsync(dst, src, sz, cudaMemcpyHostToDevice, cuStream), NCV_CUDA_ERROR);
}
else
{
ncvAssertCUDAReturn(cudaMemcpy(dst, src, sz, cudaMemcpyHostToDevice), NCV_CUDA_ERROR);
}
ncvStat = NCV_SUCCESS;
break;
case NCVMemoryTypeDevice:
if (cuStream != 0)
{
ncvAssertCUDAReturn(cudaMemcpyAsync(dst, src, sz, cudaMemcpyDeviceToDevice, cuStream), NCV_CUDA_ERROR);
}
else
{
ncvAssertCUDAReturn(cudaMemcpy(dst, src, sz, cudaMemcpyDeviceToDevice), NCV_CUDA_ERROR);
}
ncvStat = NCV_SUCCESS;
break;
default:
ncvStat = NCV_MEM_RESIDENCE_ERROR;
}
break;
default:
ncvStat = NCV_MEM_RESIDENCE_ERROR;
}
return ncvStat;
}
NCVStatus memSegCopyHelper2D(void *dst, Ncv32u dstPitch, NCVMemoryType dstType,
const void *src, Ncv32u srcPitch, NCVMemoryType srcType,
Ncv32u widthbytes, Ncv32u height, cudaStream_t cuStream)
{
NCVStatus ncvStat;
switch (dstType)
{
case NCVMemoryTypeHostPageable:
case NCVMemoryTypeHostPinned:
switch (srcType)
{
case NCVMemoryTypeHostPageable:
case NCVMemoryTypeHostPinned:
for (Ncv32u i=0; i<height; i++)
{
memcpy((char*)dst + i * dstPitch, (char*)src + i * srcPitch, widthbytes);
}
ncvStat = NCV_SUCCESS;
break;
case NCVMemoryTypeDevice:
if (cuStream != 0)
{
ncvAssertCUDAReturn(cudaMemcpy2DAsync(dst, dstPitch, src, srcPitch, widthbytes, height, cudaMemcpyDeviceToHost, cuStream), NCV_CUDA_ERROR);
}
else
{
ncvAssertCUDAReturn(cudaMemcpy2D(dst, dstPitch, src, srcPitch, widthbytes, height, cudaMemcpyDeviceToHost), NCV_CUDA_ERROR);
}
ncvStat = NCV_SUCCESS;
break;
default:
ncvStat = NCV_MEM_RESIDENCE_ERROR;
}
break;
case NCVMemoryTypeDevice:
switch (srcType)
{
case NCVMemoryTypeHostPageable:
case NCVMemoryTypeHostPinned:
if (cuStream != 0)
{
ncvAssertCUDAReturn(cudaMemcpy2DAsync(dst, dstPitch, src, srcPitch, widthbytes, height, cudaMemcpyHostToDevice, cuStream), NCV_CUDA_ERROR);
}
else
{
ncvAssertCUDAReturn(cudaMemcpy2D(dst, dstPitch, src, srcPitch, widthbytes, height, cudaMemcpyHostToDevice), NCV_CUDA_ERROR);
}
ncvStat = NCV_SUCCESS;
break;
case NCVMemoryTypeDevice:
if (cuStream != 0)
{
ncvAssertCUDAReturn(cudaMemcpy2DAsync(dst, dstPitch, src, srcPitch, widthbytes, height, cudaMemcpyDeviceToDevice, cuStream), NCV_CUDA_ERROR);
}
else
{
ncvAssertCUDAReturn(cudaMemcpy2D(dst, dstPitch, src, srcPitch, widthbytes, height, cudaMemcpyDeviceToDevice), NCV_CUDA_ERROR);
}
ncvStat = NCV_SUCCESS;
break;
default:
ncvStat = NCV_MEM_RESIDENCE_ERROR;
}
break;
default:
ncvStat = NCV_MEM_RESIDENCE_ERROR;
}
return ncvStat;
}
//===================================================================
//
// NCVMemStackAllocator class members implementation
//
//===================================================================
NCVMemStackAllocator::NCVMemStackAllocator(Ncv32u alignment_) :
_memType(NCVMemoryTypeNone),
_alignment(alignment_),
allocBegin(NULL),
begin(NULL),
end(NULL),
currentSize(0),
_maxSize(0),
bReusesMemory(false)
{
NcvBool bProperAlignment = (alignment_ & (alignment_ - 1)) == 0;
ncvAssertPrintCheck(bProperAlignment, "NCVMemStackAllocator ctor:: alignment not power of 2");
}
NCVMemStackAllocator::NCVMemStackAllocator(NCVMemoryType memT, size_t capacity, Ncv32u alignment_, void *reusePtr) :
_memType(memT),
_alignment(alignment_),
allocBegin(NULL),
currentSize(0),
_maxSize(0)
{
NcvBool bProperAlignment = (alignment_ & (alignment_ - 1)) == 0;
ncvAssertPrintCheck(bProperAlignment, "NCVMemStackAllocator ctor:: _alignment not power of 2");
ncvAssertPrintCheck(memT != NCVMemoryTypeNone, "NCVMemStackAllocator ctor:: Incorrect allocator type");
allocBegin = NULL;
if (reusePtr == NULL && capacity != 0)
{
bReusesMemory = false;
switch (memT)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaMalloc(&allocBegin, capacity), );
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaMallocHost(&allocBegin, capacity), );
break;
case NCVMemoryTypeHostPageable:
allocBegin = (Ncv8u *)malloc(capacity);
break;
default:;
}
}
else
{
bReusesMemory = true;
allocBegin = (Ncv8u *)reusePtr;
}
if (capacity == 0)
{
allocBegin = (Ncv8u *)(0x1);
}
if (!isCounting())
{
begin = allocBegin;
end = begin + capacity;
}
}
NCVMemStackAllocator::~NCVMemStackAllocator()
{
if (allocBegin != NULL)
{
ncvAssertPrintCheck(currentSize == 0, "NCVMemStackAllocator dtor:: not all objects were deallocated properly, forcing destruction");
if (!bReusesMemory && (allocBegin != (Ncv8u *)(0x1)))
{
switch (_memType)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaFree(allocBegin), );
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaFreeHost(allocBegin), );
break;
case NCVMemoryTypeHostPageable:
free(allocBegin);
break;
default:;
}
}
allocBegin = NULL;
}
}
NCVStatus NCVMemStackAllocator::alloc(NCVMemSegment &seg, size_t size)
{
seg.clear();
ncvAssertReturn(isInitialized(), NCV_ALLOCATOR_BAD_ALLOC);
size = alignUp(static_cast<Ncv32u>(size), this->_alignment);
this->currentSize += size;
this->_maxSize = std::max(this->_maxSize, this->currentSize);
if (!isCounting())
{
size_t availSize = end - begin;
ncvAssertReturn(size <= availSize, NCV_ALLOCATOR_INSUFFICIENT_CAPACITY);
}
seg.begin.ptr = begin;
seg.begin.memtype = this->_memType;
seg.size = size;
begin += size;
return NCV_SUCCESS;
}
NCVStatus NCVMemStackAllocator::dealloc(NCVMemSegment &seg)
{
ncvAssertReturn(isInitialized(), NCV_ALLOCATOR_BAD_ALLOC);
ncvAssertReturn(seg.begin.memtype == this->_memType, NCV_ALLOCATOR_BAD_DEALLOC);
ncvAssertReturn(seg.begin.ptr != NULL || isCounting(), NCV_ALLOCATOR_BAD_DEALLOC);
ncvAssertReturn(seg.begin.ptr == begin - seg.size, NCV_ALLOCATOR_DEALLOC_ORDER);
currentSize -= seg.size;
begin -= seg.size;
seg.clear();
ncvAssertReturn(allocBegin <= begin, NCV_ALLOCATOR_BAD_DEALLOC);
return NCV_SUCCESS;
}
NcvBool NCVMemStackAllocator::isInitialized(void) const
{
return (((this->_alignment & (this->_alignment-1)) == 0) && isCounting()) || this->allocBegin != NULL;
}
NcvBool NCVMemStackAllocator::isCounting(void) const
{
return this->_memType == NCVMemoryTypeNone;
}
NCVMemoryType NCVMemStackAllocator::memType(void) const
{
return this->_memType;
}
Ncv32u NCVMemStackAllocator::alignment(void) const
{
return this->_alignment;
}
size_t NCVMemStackAllocator::maxSize(void) const
{
return this->_maxSize;
}
//===================================================================
//
// NCVMemNativeAllocator class members implementation
//
//===================================================================
NCVMemNativeAllocator::NCVMemNativeAllocator(NCVMemoryType memT, Ncv32u alignment_) :
_memType(memT),
_alignment(alignment_),
currentSize(0),
_maxSize(0)
{
ncvAssertPrintReturn(memT != NCVMemoryTypeNone, "NCVMemNativeAllocator ctor:: counting not permitted for this allocator type", );
}
NCVMemNativeAllocator::~NCVMemNativeAllocator()
{
ncvAssertPrintCheck(currentSize == 0, "NCVMemNativeAllocator dtor:: detected memory leak");
}
NCVStatus NCVMemNativeAllocator::alloc(NCVMemSegment &seg, size_t size)
{
seg.clear();
ncvAssertReturn(isInitialized(), NCV_ALLOCATOR_BAD_ALLOC);
switch (this->_memType)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaMalloc(&seg.begin.ptr, size), NCV_CUDA_ERROR);
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaMallocHost(&seg.begin.ptr, size), NCV_CUDA_ERROR);
break;
case NCVMemoryTypeHostPageable:
seg.begin.ptr = (Ncv8u *)malloc(size);
break;
default:;
}
this->currentSize += alignUp(static_cast<Ncv32u>(size), this->_alignment);
this->_maxSize = std::max(this->_maxSize, this->currentSize);
seg.begin.memtype = this->_memType;
seg.size = size;
return NCV_SUCCESS;
}
NCVStatus NCVMemNativeAllocator::dealloc(NCVMemSegment &seg)
{
ncvAssertReturn(isInitialized(), NCV_ALLOCATOR_BAD_ALLOC);
ncvAssertReturn(seg.begin.memtype == this->_memType, NCV_ALLOCATOR_BAD_DEALLOC);
ncvAssertReturn(seg.begin.ptr != NULL, NCV_ALLOCATOR_BAD_DEALLOC);
ncvAssertReturn(currentSize >= alignUp(static_cast<Ncv32u>(seg.size), this->_alignment), NCV_ALLOCATOR_BAD_DEALLOC);
currentSize -= alignUp(static_cast<Ncv32u>(seg.size), this->_alignment);
switch (this->_memType)
{
case NCVMemoryTypeDevice:
ncvAssertCUDAReturn(cudaFree(seg.begin.ptr), NCV_CUDA_ERROR);
break;
case NCVMemoryTypeHostPinned:
ncvAssertCUDAReturn(cudaFreeHost(seg.begin.ptr), NCV_CUDA_ERROR);
break;
case NCVMemoryTypeHostPageable:
free(seg.begin.ptr);
break;
default:;
}
seg.clear();
return NCV_SUCCESS;
}
NcvBool NCVMemNativeAllocator::isInitialized(void) const
{
return (this->_alignment != 0);
}
NcvBool NCVMemNativeAllocator::isCounting(void) const
{
return false;
}
NCVMemoryType NCVMemNativeAllocator::memType(void) const
{
return this->_memType;
}
Ncv32u NCVMemNativeAllocator::alignment(void) const
{
return this->_alignment;
}
size_t NCVMemNativeAllocator::maxSize(void) const
{
return this->_maxSize;
}
//===================================================================
//
// Time and timer routines
//
//===================================================================
typedef struct _NcvTimeMoment NcvTimeMoment;
#if defined(_WIN32) || defined(_WIN64)
#include <Windows.h>
typedef struct _NcvTimeMoment
{
LONGLONG moment, freq;
} NcvTimeMoment;
static void _ncvQueryMoment(NcvTimeMoment *t)
{
QueryPerformanceFrequency((LARGE_INTEGER *)&(t->freq));
QueryPerformanceCounter((LARGE_INTEGER *)&(t->moment));
}
double _ncvMomentToMicroseconds(NcvTimeMoment *t)
{
return 1000000.0 * t->moment / t->freq;
}
double _ncvMomentsDiffToMicroseconds(NcvTimeMoment *t1, NcvTimeMoment *t2)
{
return 1000000.0 * 2 * ((t2->moment) - (t1->moment)) / (t1->freq + t2->freq);
}
double _ncvMomentsDiffToMilliseconds(NcvTimeMoment *t1, NcvTimeMoment *t2)
{
return 1000.0 * 2 * ((t2->moment) - (t1->moment)) / (t1->freq + t2->freq);
}
#elif defined(__GNUC__)
#include <sys/time.h>
typedef struct _NcvTimeMoment
{
struct timeval tv;
struct timezone tz;
} NcvTimeMoment;
void _ncvQueryMoment(NcvTimeMoment *t)
{
gettimeofday(& t->tv, & t->tz);
}
double _ncvMomentToMicroseconds(NcvTimeMoment *t)
{
return 1000000.0 * t->tv.tv_sec + (double)t->tv.tv_usec;
}
double _ncvMomentsDiffToMicroseconds(NcvTimeMoment *t1, NcvTimeMoment *t2)
{
return (((double)t2->tv.tv_sec - (double)t1->tv.tv_sec) * 1000000 + (double)t2->tv.tv_usec - (double)t1->tv.tv_usec);
}
double _ncvMomentsDiffToMilliseconds(NcvTimeMoment *t1, NcvTimeMoment *t2)
{
return ((double)t2->tv.tv_sec - (double)t1->tv.tv_sec) * 1000;
}
#endif //#if defined(_WIN32) || defined(_WIN64)
struct _NcvTimer
{
NcvTimeMoment t1, t2;
};
NcvTimer ncvStartTimer(void)
{
struct _NcvTimer *t;
t = (struct _NcvTimer *)malloc(sizeof(struct _NcvTimer));
_ncvQueryMoment(&t->t1);
return t;
}
double ncvEndQueryTimerUs(NcvTimer t)
{
double res;
_ncvQueryMoment(&t->t2);
res = _ncvMomentsDiffToMicroseconds(&t->t1, &t->t2);
free(t);
return res;
}
double ncvEndQueryTimerMs(NcvTimer t)
{
double res;
_ncvQueryMoment(&t->t2);
res = _ncvMomentsDiffToMilliseconds(&t->t1, &t->t2);
free(t);
return res;
}
//===================================================================
//
// Operations with rectangles
//
//===================================================================
struct RectConvert
{
cv::Rect operator()(const NcvRect32u& nr) const { return cv::Rect(nr.x, nr.y, nr.width, nr.height); }
NcvRect32u operator()(const cv::Rect& nr) const
{
NcvRect32u rect;
rect.x = nr.x;
rect.y = nr.y;
rect.width = nr.width;
rect.height = nr.height;
return rect;
}
};
static void groupRectangles(std::vector<NcvRect32u> &hypotheses, int groupThreshold, double eps, std::vector<Ncv32u> *weights)
{
#ifndef HAVE_OPENCV_XOBJDETECT
CV_UNUSED(hypotheses);
CV_UNUSED(groupThreshold);
CV_UNUSED(eps);
CV_UNUSED(weights);
CV_Error(cv::Error::StsNotImplemented, "This functionality requires xobjdetect module");
#else
std::vector<cv::Rect> rects(hypotheses.size());
std::transform(hypotheses.begin(), hypotheses.end(), rects.begin(), RectConvert());
if (weights)
{
std::vector<int> weights_int;
weights_int.assign(weights->begin(), weights->end());
cv::groupRectangles(rects, weights_int, groupThreshold, eps);
}
else
{
cv::groupRectangles(rects, groupThreshold, eps);
}
std::transform(rects.begin(), rects.end(), hypotheses.begin(), RectConvert());
hypotheses.resize(rects.size());
#endif
}
NCVStatus ncvGroupRectangles_host(NCVVector<NcvRect32u> &hypotheses,
Ncv32u &numHypotheses,
Ncv32u minNeighbors,
Ncv32f intersectEps,
NCVVector<Ncv32u> *hypothesesWeights)
{
ncvAssertReturn(hypotheses.memType() == NCVMemoryTypeHostPageable ||
hypotheses.memType() == NCVMemoryTypeHostPinned, NCV_MEM_RESIDENCE_ERROR);
if (hypothesesWeights != NULL)
{
ncvAssertReturn(hypothesesWeights->memType() == NCVMemoryTypeHostPageable ||
hypothesesWeights->memType() == NCVMemoryTypeHostPinned, NCV_MEM_RESIDENCE_ERROR);
}
if (numHypotheses == 0)
{
return NCV_SUCCESS;
}
std::vector<NcvRect32u> rects(numHypotheses);
memcpy(&rects[0], hypotheses.ptr(), numHypotheses * sizeof(NcvRect32u));
std::vector<Ncv32u> weights;
if (hypothesesWeights != NULL)
{
groupRectangles(rects, minNeighbors, intersectEps, &weights);
}
else
{
groupRectangles(rects, minNeighbors, intersectEps, NULL);
}
numHypotheses = (Ncv32u)rects.size();
if (numHypotheses > 0)
{
memcpy(hypotheses.ptr(), &rects[0], numHypotheses * sizeof(NcvRect32u));
}
if (hypothesesWeights != NULL)
{
memcpy(hypothesesWeights->ptr(), &weights[0], numHypotheses * sizeof(Ncv32u));
}
return NCV_SUCCESS;
}
template <class T>
static NCVStatus drawRectsWrapperHost(T *h_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *h_rects,
Ncv32u numRects,
T color)
{
ncvAssertReturn(h_dst != NULL && h_rects != NULL, NCV_NULL_PTR);
ncvAssertReturn(dstWidth > 0 && dstHeight > 0, NCV_DIMENSIONS_INVALID);
ncvAssertReturn(dstStride >= dstWidth, NCV_INVALID_STEP);
ncvAssertReturn(numRects != 0, NCV_SUCCESS);
ncvAssertReturn(numRects <= dstWidth * dstHeight, NCV_DIMENSIONS_INVALID);
for (Ncv32u i=0; i<numRects; i++)
{
NcvRect32u rect = h_rects[i];
if (rect.x < dstWidth)
{
for (Ncv32u each=rect.y; each<rect.y+rect.height && each<dstHeight; each++)
{
h_dst[each*dstStride+rect.x] = color;
}
}
if (rect.x+rect.width-1 < dstWidth)
{
for (Ncv32u each=rect.y; each<rect.y+rect.height && each<dstHeight; each++)
{
h_dst[each*dstStride+rect.x+rect.width-1] = color;
}
}
if (rect.y < dstHeight)
{
for (Ncv32u j=rect.x; j<rect.x+rect.width && j<dstWidth; j++)
{
h_dst[rect.y*dstStride+j] = color;
}
}
if (rect.y + rect.height - 1 < dstHeight)
{
for (Ncv32u j=rect.x; j<rect.x+rect.width && j<dstWidth; j++)
{
h_dst[(rect.y+rect.height-1)*dstStride+j] = color;
}
}
}
return NCV_SUCCESS;
}
NCVStatus ncvDrawRects_8u_host(Ncv8u *h_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *h_rects,
Ncv32u numRects,
Ncv8u color)
{
return drawRectsWrapperHost(h_dst, dstStride, dstWidth, dstHeight, h_rects, numRects, color);
}
NCVStatus ncvDrawRects_32u_host(Ncv32u *h_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *h_rects,
Ncv32u numRects,
Ncv32u color)
{
return drawRectsWrapperHost(h_dst, dstStride, dstWidth, dstHeight, h_rects, numRects, color);
}
#endif
modules/cudalegacy/src/bm.cpp
+204
View File
@@ -0,0 +1,204 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined HAVE_CUDA || defined(CUDA_DISABLER)
void cv::cuda::calcOpticalFlowBM(const GpuMat&, const GpuMat&, Size, Size, Size, bool, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
#else // HAVE_CUDA
namespace optflowbm
{
void calc(PtrStepSzb prev, PtrStepSzb curr, PtrStepSzf velx, PtrStepSzf vely, int2 blockSize, int2 shiftSize, bool usePrevious,
int maxX, int maxY, int acceptLevel, int escapeLevel, const short2* ss, int ssCount, cudaStream_t stream);
}
void cv::cuda::calcOpticalFlowBM(const GpuMat& prev, const GpuMat& curr, Size blockSize, Size shiftSize, Size maxRange, bool usePrevious, GpuMat& velx, GpuMat& vely, GpuMat& buf, Stream& st)
{
CV_Assert( prev.type() == CV_8UC1 );
CV_Assert( curr.size() == prev.size() && curr.type() == prev.type() );
const Size velSize((prev.cols - blockSize.width + shiftSize.width) / shiftSize.width,
(prev.rows - blockSize.height + shiftSize.height) / shiftSize.height);
velx.create(velSize, CV_32FC1);
vely.create(velSize, CV_32FC1);
// scanning scheme coordinates
std::vector<short2> ss((2 * maxRange.width + 1) * (2 * maxRange.height + 1));
int ssCount = 0;
// Calculate scanning scheme
const int minCount = std::min(maxRange.width, maxRange.height);
// use spiral search pattern
//
// 9 10 11 12
// 8 1 2 13
// 7 * 3 14
// 6 5 4 15
//... 20 19 18 17
//
for (int i = 0; i < minCount; ++i)
{
// four cycles along sides
int x = -i - 1, y = x;
// upper side
for (int j = -i; j <= i + 1; ++j, ++ssCount)
{
ss[ssCount].x = (short) ++x;
ss[ssCount].y = (short) y;
}
// right side
for (int j = -i; j <= i + 1; ++j, ++ssCount)
{
ss[ssCount].x = (short) x;
ss[ssCount].y = (short) ++y;
}
// bottom side
for (int j = -i; j <= i + 1; ++j, ++ssCount)
{
ss[ssCount].x = (short) --x;
ss[ssCount].y = (short) y;
}
// left side
for (int j = -i; j <= i + 1; ++j, ++ssCount)
{
ss[ssCount].x = (short) x;
ss[ssCount].y = (short) --y;
}
}
// the rest part
if (maxRange.width < maxRange.height)
{
const int xleft = -minCount;
// cycle by neighbor rings
for (int i = minCount; i < maxRange.height; ++i)
{
// two cycles by x
int y = -(i + 1);
int x = xleft;
// upper side
for (int j = -maxRange.width; j <= maxRange.width; ++j, ++ssCount, ++x)
{
ss[ssCount].x = (short) x;
ss[ssCount].y = (short) y;
}
x = xleft;
y = -y;
// bottom side
for (int j = -maxRange.width; j <= maxRange.width; ++j, ++ssCount, ++x)
{
ss[ssCount].x = (short) x;
ss[ssCount].y = (short) y;
}
}
}
else if (maxRange.width > maxRange.height)
{
const int yupper = -minCount;
// cycle by neighbor rings
for (int i = minCount; i < maxRange.width; ++i)
{
// two cycles by y
int x = -(i + 1);
int y = yupper;
// left side
for (int j = -maxRange.height; j <= maxRange.height; ++j, ++ssCount, ++y)
{
ss[ssCount].x = (short) x;
ss[ssCount].y = (short) y;
}
y = yupper;
x = -x;
// right side
for (int j = -maxRange.height; j <= maxRange.height; ++j, ++ssCount, ++y)
{
ss[ssCount].x = (short) x;
ss[ssCount].y = (short) y;
}
}
}
const cudaStream_t stream = StreamAccessor::getStream(st);
ensureSizeIsEnough(1, ssCount, CV_16SC2, buf);
if (stream == 0)
cudaSafeCall( cudaMemcpy(buf.data, &ss[0], ssCount * sizeof(short2), cudaMemcpyHostToDevice) );
else
cudaSafeCall( cudaMemcpyAsync(buf.data, &ss[0], ssCount * sizeof(short2), cudaMemcpyHostToDevice, stream) );
const int maxX = prev.cols - blockSize.width;
const int maxY = prev.rows - blockSize.height;
const int SMALL_DIFF = 2;
const int BIG_DIFF = 128;
const int blSize = blockSize.area();
const int acceptLevel = blSize * SMALL_DIFF;
const int escapeLevel = blSize * BIG_DIFF;
optflowbm::calc(prev, curr, velx, vely,
make_int2(blockSize.width, blockSize.height), make_int2(shiftSize.width, shiftSize.height), usePrevious,
maxX, maxY, acceptLevel, escapeLevel, buf.ptr<short2>(), ssCount, stream);
}
#endif // HAVE_CUDA
modules/cudalegacy/src/bm_fast.cpp
+90
View File
@@ -0,0 +1,90 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined HAVE_CUDA || !defined(HAVE_OPENCV_CUDAARITHM) || defined(CUDA_DISABLER)
void cv::cuda::FastOpticalFlowBM::operator ()(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, int, int, Stream&) { throw_no_cuda(); }
#else // HAVE_CUDA
namespace optflowbm_fast
{
void get_buffer_size(int src_cols, int src_rows, int search_window, int block_window, int& buffer_cols, int& buffer_rows);
template <typename T>
void calc(PtrStepSzb I0, PtrStepSzb I1, PtrStepSzf velx, PtrStepSzf vely, PtrStepi buffer, int search_window, int block_window, cudaStream_t stream);
}
void cv::cuda::FastOpticalFlowBM::operator ()(const GpuMat& I0, const GpuMat& I1, GpuMat& flowx, GpuMat& flowy, int search_window, int block_window, Stream& stream)
{
CV_Assert( I0.type() == CV_8UC1 );
CV_Assert( I1.size() == I0.size() && I1.type() == I0.type() );
int border_size = search_window / 2 + block_window / 2;
Size esize = I0.size() + Size(border_size, border_size) * 2;
ensureSizeIsEnough(esize, I0.type(), extended_I0);
ensureSizeIsEnough(esize, I0.type(), extended_I1);
cuda::copyMakeBorder(I0, extended_I0, border_size, border_size, border_size, border_size, cv::BORDER_DEFAULT, Scalar(), stream);
cuda::copyMakeBorder(I1, extended_I1, border_size, border_size, border_size, border_size, cv::BORDER_DEFAULT, Scalar(), stream);
GpuMat I0_hdr = extended_I0(Rect(Point2i(border_size, border_size), I0.size()));
GpuMat I1_hdr = extended_I1(Rect(Point2i(border_size, border_size), I0.size()));
int bcols, brows;
optflowbm_fast::get_buffer_size(I0.cols, I0.rows, search_window, block_window, bcols, brows);
ensureSizeIsEnough(brows, bcols, CV_32SC1, buffer);
flowx.create(I0.size(), CV_32FC1);
flowy.create(I0.size(), CV_32FC1);
optflowbm_fast::calc<uchar>(I0_hdr, I1_hdr, flowx, flowy, buffer, search_window, block_window, StreamAccessor::getStream(stream));
}
#endif // HAVE_CUDA
modules/cudalegacy/src/calib3d.cpp
+292
View File
@@ -0,0 +1,292 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined HAVE_CUDA || !defined HAVE_OPENCV_3D || defined(CUDA_DISABLER)
void cv::cuda::transformPoints(const GpuMat&, const Mat&, const Mat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::cuda::projectPoints(const GpuMat&, const Mat&, const Mat&, const Mat&, const Mat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::cuda::solvePnPRansac(const Mat&, const Mat&, const Mat&, const Mat&, Mat&, Mat&, bool, int, float, int, std::vector<int>*) { throw_no_cuda(); }
#else
namespace cv { namespace cuda { namespace device
{
namespace transform_points
{
void call(const PtrStepSz<float3> src, const float* rot, const float* transl, PtrStepSz<float3> dst, cudaStream_t stream);
}
namespace project_points
{
void call(const PtrStepSz<float3> src, const float* rot, const float* transl, const float* proj, PtrStepSz<float2> dst, cudaStream_t stream);
}
namespace solve_pnp_ransac
{
int maxNumIters();
void computeHypothesisScores(
const int num_hypotheses, const int num_points, const float* rot_matrices,
const float3* transl_vectors, const float3* object, const float2* image,
const float dist_threshold, int* hypothesis_scores);
}
}}}
using namespace ::cv::cuda::device;
namespace
{
void transformPointsCaller(const GpuMat& src, const Mat& rvec, const Mat& tvec, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src.rows == 1 && src.cols > 0 && src.type() == CV_32FC3);
CV_Assert(rvec.size() == Size(3, 1) && rvec.type() == CV_32F);
CV_Assert(tvec.size() == Size(3, 1) && tvec.type() == CV_32F);
// Convert rotation vector into matrix
Mat rot;
Rodrigues(rvec, rot);
dst.create(src.size(), src.type());
transform_points::call(src, rot.ptr<float>(), tvec.ptr<float>(), dst, stream);
}
}
void cv::cuda::transformPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec, GpuMat& dst, Stream& stream)
{
transformPointsCaller(src, rvec, tvec, dst, StreamAccessor::getStream(stream));
}
namespace
{
void projectPointsCaller(const GpuMat& src, const Mat& rvec, const Mat& tvec, const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src.rows == 1 && src.cols > 0 && src.type() == CV_32FC3);
CV_Assert(rvec.size() == Size(3, 1) && rvec.type() == CV_32F);
CV_Assert(tvec.size() == Size(3, 1) && tvec.type() == CV_32F);
CV_Assert(camera_mat.size() == Size(3, 3) && camera_mat.type() == CV_32F);
CV_Assert(dist_coef.empty()); // Undistortion isn't supported
// Convert rotation vector into matrix
Mat rot;
Rodrigues(rvec, rot);
dst.create(src.size(), CV_32FC2);
project_points::call(src, rot.ptr<float>(), tvec.ptr<float>(), camera_mat.ptr<float>(), dst,stream);
}
}
void cv::cuda::projectPoints(const GpuMat& src, const Mat& rvec, const Mat& tvec, const Mat& camera_mat, const Mat& dist_coef, GpuMat& dst, Stream& stream)
{
projectPointsCaller(src, rvec, tvec, camera_mat, dist_coef, dst, StreamAccessor::getStream(stream));
}
namespace
{
// Selects subset_size random different points from [0, num_points - 1] range
void selectRandom(int subset_size, int num_points, std::vector<int>& subset)
{
subset.resize(subset_size);
for (int i = 0; i < subset_size; ++i)
{
bool was;
do
{
subset[i] = rand() % num_points;
was = false;
for (int j = 0; j < i; ++j)
if (subset[j] == subset[i])
{
was = true;
break;
}
} while (was);
}
}
// Computes rotation, translation pair for small subsets if the input data
class TransformHypothesesGenerator : public ParallelLoopBody
{
public:
TransformHypothesesGenerator(const Mat& object_, const Mat& image_, const Mat& dist_coef_,
const Mat& camera_mat_, int num_points_, int subset_size_,
Mat rot_matrices_, Mat transl_vectors_)
: object(&object_), image(&image_), dist_coef(&dist_coef_), camera_mat(&camera_mat_),
num_points(num_points_), subset_size(subset_size_), rot_matrices(rot_matrices_),
transl_vectors(transl_vectors_) {}
void operator()(const Range& range) const CV_OVERRIDE
{
// Input data for generation of the current hypothesis
std::vector<int> subset_indices(subset_size);
Mat_<Point3f> object_subset(1, subset_size);
Mat_<Point2f> image_subset(1, subset_size);
// Current hypothesis data
Mat rot_vec(1, 3, CV_64F);
Mat rot_mat(3, 3, CV_64F);
Mat transl_vec(1, 3, CV_64F);
for (int iter = range.start; iter < range.end; ++iter)
{
selectRandom(subset_size, num_points, subset_indices);
for (int i = 0; i < subset_size; ++i)
{
object_subset(0, i) = object->at<Point3f>(subset_indices[i]);
image_subset(0, i) = image->at<Point2f>(subset_indices[i]);
}
solvePnP(object_subset, image_subset, *camera_mat, *dist_coef, rot_vec, transl_vec, false, SOLVEPNP_EPNP);
// Remember translation vector
Mat transl_vec_ = transl_vectors.colRange(iter * 3, (iter + 1) * 3);
transl_vec = transl_vec.reshape(0, 1);
transl_vec.convertTo(transl_vec_, CV_32F);
// Remember rotation matrix
Rodrigues(rot_vec, rot_mat);
Mat rot_mat_ = rot_matrices.colRange(iter * 9, (iter + 1) * 9).reshape(0, 3);
rot_mat.convertTo(rot_mat_, CV_32F);
}
}
const Mat* object;
const Mat* image;
const Mat* dist_coef;
const Mat* camera_mat;
int num_points;
int subset_size;
// Hypotheses storage (global)
Mat rot_matrices;
Mat transl_vectors;
};
}
void cv::cuda::solvePnPRansac(const Mat& object, const Mat& image, const Mat& camera_mat,
const Mat& dist_coef, Mat& rvec, Mat& tvec, bool use_extrinsic_guess,
int num_iters, float max_dist, int min_inlier_count,
std::vector<int>* inliers)
{
CV_UNUSED(min_inlier_count);
CV_Assert(object.rows == 1 && object.cols > 0 && object.type() == CV_32FC3);
CV_Assert(image.rows == 1 && image.cols > 0 && image.type() == CV_32FC2);
CV_Assert(object.cols == image.cols);
CV_Assert(camera_mat.size() == Size(3, 3) && camera_mat.type() == CV_32F);
CV_Assert(!use_extrinsic_guess); // We don't support initial guess for now
CV_Assert(num_iters <= solve_pnp_ransac::maxNumIters());
const int subset_size = 4;
const int num_points = object.cols;
CV_Assert(num_points >= subset_size);
// Unapply distortion and intrinsic camera transformations
Mat eye_camera_mat = Mat::eye(3, 3, CV_32F);
Mat empty_dist_coef;
Mat image_normalized;
undistortPoints(image, image_normalized, camera_mat, dist_coef, Mat(), eye_camera_mat);
// Hypotheses storage (global)
Mat rot_matrices(1, num_iters * 9, CV_32F);
Mat transl_vectors(1, num_iters * 3, CV_32F);
// Generate set of hypotheses using small subsets of the input data
TransformHypothesesGenerator body(object, image_normalized, empty_dist_coef, eye_camera_mat,
num_points, subset_size, rot_matrices, transl_vectors);
parallel_for_(Range(0, num_iters), body);
// Compute scores (i.e. number of inliers) for each hypothesis
GpuMat d_object(object);
GpuMat d_image_normalized(image_normalized);
GpuMat d_hypothesis_scores(1, num_iters, CV_32S);
solve_pnp_ransac::computeHypothesisScores(
num_iters, num_points, rot_matrices.ptr<float>(), transl_vectors.ptr<float3>(),
d_object.ptr<float3>(), d_image_normalized.ptr<float2>(), max_dist * max_dist,
d_hypothesis_scores.ptr<int>());
// Find the best hypothesis index
Point best_idx;
double best_score;
cuda::minMaxLoc(d_hypothesis_scores, NULL, &best_score, NULL, &best_idx);
int num_inliers = static_cast<int>(best_score);
// Extract the best hypothesis data
Mat rot_mat = rot_matrices.colRange(best_idx.x * 9, (best_idx.x + 1) * 9).reshape(0, 3);
Rodrigues(rot_mat, rvec);
rvec = rvec.reshape(0, 1);
tvec = transl_vectors.colRange(best_idx.x * 3, (best_idx.x + 1) * 3).clone();
tvec = tvec.reshape(0, 1);
// Build vector of inlier indices
if (inliers != NULL)
{
inliers->clear();
inliers->reserve(num_inliers);
Point3f p, p_transf;
Point2f p_proj;
const float* rot = rot_mat.ptr<float>();
const float* transl = tvec.ptr<float>();
for (int i = 0; i < num_points; ++i)
{
p = object.at<Point3f>(0, i);
p_transf.x = rot[0] * p.x + rot[1] * p.y + rot[2] * p.z + transl[0];
p_transf.y = rot[3] * p.x + rot[4] * p.y + rot[5] * p.z + transl[1];
p_transf.z = rot[6] * p.x + rot[7] * p.y + rot[8] * p.z + transl[2];
p_proj.x = p_transf.x / p_transf.z;
p_proj.y = p_transf.y / p_transf.z;
if (norm(p_proj - image_normalized.at<Point2f>(i)) < max_dist)
inliers->push_back(i);
}
}
}
#endif
modules/cudalegacy/src/cuda/NCV.cu
+180
View File
@@ -0,0 +1,180 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include <iostream>
#include <vector>
#include "opencv2/cudalegacy/NCV.hpp"
//===================================================================
//
// Operations with rectangles
//
//===================================================================
const Ncv32u NUMTHREADS_DRAWRECTS = 32;
const Ncv32u NUMTHREADS_DRAWRECTS_LOG2 = 5;
template <class T>
__global__ void drawRects(T *d_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *d_rects,
Ncv32u numRects,
T color)
{
Ncv32u blockId = blockIdx.y * 65535 + blockIdx.x;
if (blockId > numRects * 4)
{
return;
}
NcvRect32u curRect = d_rects[blockId >> 2];
NcvBool bVertical = blockId & 0x1;
NcvBool bTopLeft = blockId & 0x2;
Ncv32u pt0x, pt0y;
if (bVertical)
{
Ncv32u numChunks = (curRect.height + NUMTHREADS_DRAWRECTS - 1) >> NUMTHREADS_DRAWRECTS_LOG2;
pt0x = bTopLeft ? curRect.x : curRect.x + curRect.width - 1;
pt0y = curRect.y;
if (pt0x < dstWidth)
{
for (Ncv32u chunkId = 0; chunkId < numChunks; chunkId++)
{
Ncv32u ptY = pt0y + chunkId * NUMTHREADS_DRAWRECTS + threadIdx.x;
if (ptY < pt0y + curRect.height && ptY < dstHeight)
{
d_dst[ptY * dstStride + pt0x] = color;
}
}
}
}
else
{
Ncv32u numChunks = (curRect.width + NUMTHREADS_DRAWRECTS - 1) >> NUMTHREADS_DRAWRECTS_LOG2;
pt0x = curRect.x;
pt0y = bTopLeft ? curRect.y : curRect.y + curRect.height - 1;
if (pt0y < dstHeight)
{
for (Ncv32u chunkId = 0; chunkId < numChunks; chunkId++)
{
Ncv32u ptX = pt0x + chunkId * NUMTHREADS_DRAWRECTS + threadIdx.x;
if (ptX < pt0x + curRect.width && ptX < dstWidth)
{
d_dst[pt0y * dstStride + ptX] = color;
}
}
}
}
}
template <class T>
static NCVStatus drawRectsWrapperDevice(T *d_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *d_rects,
Ncv32u numRects,
T color,
cudaStream_t cuStream)
{
CV_UNUSED(cuStream);
ncvAssertReturn(d_dst != NULL && d_rects != NULL, NCV_NULL_PTR);
ncvAssertReturn(dstWidth > 0 && dstHeight > 0, NCV_DIMENSIONS_INVALID);
ncvAssertReturn(dstStride >= dstWidth, NCV_INVALID_STEP);
ncvAssertReturn(numRects <= dstWidth * dstHeight, NCV_DIMENSIONS_INVALID);
if (numRects == 0)
{
return NCV_SUCCESS;
}
dim3 grid(numRects * 4);
dim3 block(NUMTHREADS_DRAWRECTS);
if (grid.x > 65535)
{
grid.y = (grid.x + 65534) / 65535;
grid.x = 65535;
}
drawRects<T><<<grid, block>>>(d_dst, dstStride, dstWidth, dstHeight, d_rects, numRects, color);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
return NCV_SUCCESS;
}
NCVStatus ncvDrawRects_8u_device(Ncv8u *d_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *d_rects,
Ncv32u numRects,
Ncv8u color,
cudaStream_t cuStream)
{
return drawRectsWrapperDevice(d_dst, dstStride, dstWidth, dstHeight, d_rects, numRects, color, cuStream);
}
NCVStatus ncvDrawRects_32u_device(Ncv32u *d_dst,
Ncv32u dstStride,
Ncv32u dstWidth,
Ncv32u dstHeight,
NcvRect32u *d_rects,
Ncv32u numRects,
Ncv32u color,
cudaStream_t cuStream)
{
return drawRectsWrapperDevice(d_dst, dstStride, dstWidth, dstHeight, d_rects, numRects, color, cuStream);
}
modules/cudalegacy/src/cuda/NCVAlg.hpp
+155
View File
@@ -0,0 +1,155 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncv_alg_hpp_
#define _ncv_alg_hpp_
#include "opencv2/cudalegacy/NCV.hpp"
template <class T>
static void swap(T &p1, T &p2)
{
T tmp = p1;
p1 = p2;
p2 = tmp;
}
template<typename T>
static T divUp(T a, T b)
{
return (a + b - 1) / b;
}
template<typename T>
struct functorAddValues
{
static __device__ __inline__ void assign(volatile T *dst, volatile T *src)
{
//Works only for integral types. If you see compiler error here, then you have to specify how to copy your object as a set of integral fields.
*dst = *src;
}
static __device__ __inline__ void reduce(volatile T &in1out, const volatile T &in2)
{
in1out += in2;
}
};
template<typename T>
struct functorMinValues
{
static __device__ __inline__ void assign(volatile T *dst, volatile T *src)
{
//Works only for integral types. If you see compiler error here, then you have to specify how to copy your object as a set of integral fields.
*dst = *src;
}
static __device__ __inline__ void reduce(volatile T &in1out, const volatile T &in2)
{
in1out = in1out > in2 ? in2 : in1out;
}
};
template<typename T>
struct functorMaxValues
{
static __device__ __inline__ void assign(volatile T *dst, volatile T *src)
{
//Works only for integral types. If you see compiler error here, then you have to specify how to copy your object as a set of integral fields.
*dst = *src;
}
static __device__ __inline__ void reduce(volatile T &in1out, const volatile T &in2)
{
in1out = in1out > in2 ? in1out : in2;
}
};
template<typename Tdata, class Tfunc, Ncv32u nThreads>
static __device__ Tdata subReduce(Tdata threadElem)
{
Tfunc functor;
__shared__ Tdata _reduceArr[nThreads];
volatile Tdata *reduceArr = _reduceArr;
functor.assign(reduceArr + threadIdx.x, &threadElem);
__syncthreads();
if (nThreads >= 256 && threadIdx.x < 128)
{
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 128]);
}
__syncthreads();
if (nThreads >= 128 && threadIdx.x < 64)
{
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 64]);
}
__syncthreads();
if (threadIdx.x < 32)
{
if (nThreads >= 64)
{
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 32]);
}
if (nThreads >= 32 && threadIdx.x < 16)
{
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 16]);
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 8]);
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 4]);
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 2]);
functor.reduce(reduceArr[threadIdx.x], reduceArr[threadIdx.x + 1]);
}
}
__syncthreads();
Tdata reduceRes;
functor.assign(&reduceRes, reduceArr);
return reduceRes;
}
#endif //_ncv_alg_hpp_
modules/cudalegacy/src/cuda/NCVBroxOpticalFlow.cu
+985
View File
@@ -0,0 +1,985 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
////////////////////////////////////////////////////////////////////////////////
//
// NVIDIA CUDA implementation of Brox et al Optical Flow algorithm
//
// Algorithm is explained in the original paper:
// T. Brox, A. Bruhn, N. Papenberg, J. Weickert:
// High accuracy optical flow estimation based on a theory for warping.
// ECCV 2004.
//
// Implementation by Mikhail Smirnov
// email: msmirnov@nvidia.com, devsupport@nvidia.com
//
// Credits for help with the code to:
// Alexey Mendelenko, Anton Obukhov, and Alexander Kharlamov.
//
////////////////////////////////////////////////////////////////////////////////
#include <iostream>
#include <vector>
#include <memory>
#include "opencv2/core/cuda/utility.hpp"
#include "opencv2/cudalegacy/NPP_staging.hpp"
#include "opencv2/cudalegacy/NCVBroxOpticalFlow.hpp"
#include <opencv2/cudev/ptr2d/texture.hpp>
typedef NCVVectorAlloc<Ncv32f> FloatVector;
typedef cv::cudev::TexturePtr<float> Ptr2D;
typedef cv::cudev::Texture<float> Texture;
/////////////////////////////////////////////////////////////////////////////////////////
// Implementation specific constants
/////////////////////////////////////////////////////////////////////////////////////////
__device__ const float eps2 = 1e-6f;
/////////////////////////////////////////////////////////////////////////////////////////
// Additional defines
/////////////////////////////////////////////////////////////////////////////////////////
// rounded up division
inline int iDivUp(int a, int b)
{
return (a + b - 1)/b;
}
/////////////////////////////////////////////////////////////////////////////////////////
// SUPPLEMENTARY FUNCTIONS
/////////////////////////////////////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////
/// \brief performs pointwise summation of two vectors stored in device memory
/// \param d_res - pointer to resulting vector (device memory)
/// \param d_op1 - term #1 (device memory)
/// \param d_op2 - term #2 (device memory)
/// \param len - vector size
///////////////////////////////////////////////////////////////////////////////
__global__ void pointwise_add(float *d_res, const float *d_op1, const float *d_op2, const int len)
{
const int pos = blockIdx.x*blockDim.x + threadIdx.x;
if(pos >= len) return;
d_res[pos] = d_op1[pos] + d_op2[pos];
}
///////////////////////////////////////////////////////////////////////////////
/// \brief wrapper for summation kernel.
/// Computes \b op1 + \b op2 and stores result to \b res
/// \param res array, containing op1 + op2 (device memory)
/// \param op1 term #1 (device memory)
/// \param op2 term #2 (device memory)
/// \param count vector size
///////////////////////////////////////////////////////////////////////////////
static void add(float *res, const float *op1, const float *op2, const int count, cudaStream_t stream)
{
dim3 threads(256);
dim3 blocks(iDivUp(count, threads.x));
pointwise_add<<<blocks, threads, 0, stream>>>(res, op1, op2, count);
}
///////////////////////////////////////////////////////////////////////////////
/// \brief wrapper for summation kernel.
/// Increments \b res by \b rhs
/// \param res initial vector, will be replaced with result (device memory)
/// \param rhs increment (device memory)
/// \param count vector size
///////////////////////////////////////////////////////////////////////////////
static void add(float *res, const float *rhs, const int count, cudaStream_t stream)
{
add(res, res, rhs, count, stream);
}
///////////////////////////////////////////////////////////////////////////////
/// \brief kernel for scaling vector by scalar
/// \param d_res scaled vector (device memory)
/// \param d_src source vector (device memory)
/// \param scale scalar to scale by
/// \param len vector size (number of elements)
///////////////////////////////////////////////////////////////////////////////
__global__ void scaleVector(float *d_res, const float *d_src, float scale, const int len)
{
const int pos = blockIdx.x * blockDim.x + threadIdx.x;
if (pos >= len) return;
d_res[pos] = d_src[pos] * scale;
}
///////////////////////////////////////////////////////////////////////////////
/// \brief scale vector by scalar
///
/// kernel wrapper
/// \param d_res scaled vector (device memory)
/// \param d_src source vector (device memory)
/// \param scale scalar to scale by
/// \param len vector size (number of elements)
/// \param stream CUDA stream
///////////////////////////////////////////////////////////////////////////////
static void ScaleVector(float *d_res, const float *d_src, float scale, const int len, cudaStream_t stream)
{
dim3 threads(256);
dim3 blocks(iDivUp(len, threads.x));
scaleVector<<<blocks, threads, 0, stream>>>(d_res, d_src, scale, len);
}
const int SOR_TILE_WIDTH = 32;
const int SOR_TILE_HEIGHT = 6;
const int PSOR_TILE_WIDTH = 32;
const int PSOR_TILE_HEIGHT = 6;
const int PSOR_PITCH = PSOR_TILE_WIDTH + 4;
const int PSOR_HEIGHT = PSOR_TILE_HEIGHT + 4;
///////////////////////////////////////////////////////////////////////////////
///\brief Utility function. Compute smooth term diffusivity along x axis
///\param s (out) pointer to memory location for result (diffusivity)
///\param pos (in) position within shared memory array containing \b u
///\param u (in) shared memory array containing \b u
///\param v (in) shared memory array containing \b v
///\param du (in) shared memory array containing \b du
///\param dv (in) shared memory array containing \b dv
///////////////////////////////////////////////////////////////////////////////
__forceinline__ __device__ void diffusivity_along_x(float *s, int pos, const float *u, const float *v, const float *du, const float *dv)
{
//x derivative between pixels (i,j) and (i-1,j)
const int left = pos-1;
float u_x = u[pos] + du[pos] - u[left] - du[left];
float v_x = v[pos] + dv[pos] - v[left] - dv[left];
const int up = pos + PSOR_PITCH;
const int down = pos - PSOR_PITCH;
const int up_left = up - 1;
const int down_left = down-1;
//y derivative between pixels (i,j) and (i-1,j)
float u_y = 0.25f*(u[up] + du[up] + u[up_left] + du[up_left] - u[down] - du[down] - u[down_left] - du[down_left]);
float v_y = 0.25f*(v[up] + dv[up] + v[up_left] + dv[up_left] - v[down] - dv[down] - v[down_left] - dv[down_left]);
*s = 0.5f / sqrtf(u_x*u_x + v_x*v_x + u_y*u_y + v_y*v_y + eps2);
}
///////////////////////////////////////////////////////////////////////////////
///\brief Utility function. Compute smooth term diffusivity along y axis
///\param s (out) pointer to memory location for result (diffusivity)
///\param pos (in) position within shared memory array containing \b u
///\param u (in) shared memory array containing \b u
///\param v (in) shared memory array containing \b v
///\param du (in) shared memory array containing \b du
///\param dv (in) shared memory array containing \b dv
///////////////////////////////////////////////////////////////////////////////
__forceinline__ __device__ void diffusivity_along_y(float *s, int pos, const float *u, const float *v, const float *du, const float *dv)
{
//y derivative between pixels (i,j) and (i,j-1)
const int down = pos-PSOR_PITCH;
float u_y = u[pos] + du[pos] - u[down] - du[down];
float v_y = v[pos] + dv[pos] - v[down] - dv[down];
const int right = pos + 1;
const int left = pos - 1;
const int down_right = down + 1;
const int down_left = down - 1;
//x derivative between pixels (i,j) and (i,j-1);
float u_x = 0.25f*(u[right] + u[down_right] + du[right] + du[down_right] - u[left] - u[down_left] - du[left] - du[down_left]);
float v_x = 0.25f*(v[right] + v[down_right] + dv[right] + dv[down_right] - v[left] - v[down_left] - dv[left] - dv[down_left]);
*s = 0.5f/sqrtf(u_x*u_x + v_x*v_x + u_y*u_y + v_y*v_y + eps2);
}
///////////////////////////////////////////////////////////////////////////////
///\brief Utility function. Load element of 2D global memory to shared memory
///\param smem pointer to shared memory array
///\param is shared memory array column
///\param js shared memory array row
///\param w number of columns in global memory array
///\param h number of rows in global memory array
///\param p global memory array pitch in floats
///////////////////////////////////////////////////////////////////////////////
__forceinline__ __device__ void load_array_element(Ptr2D texSrc, float *smem, int is, int js, int i, int j, int w, int h, int p)
{
//position within shared memory array
const int ijs = js * PSOR_PITCH + is;
//mirror reflection across borders
i = max(i, -i-1);
i = min(i, w-i+w-1);
j = max(j, -j-1);
j = min(j, h-j+h-1);
const int pos = j * p + i;
smem[ijs] = texSrc(pos);
}
///////////////////////////////////////////////////////////////////////////////
///\brief Utility function. Load part (tile) of 2D global memory to shared memory
///\param smem pointer to target shared memory array
///\param ig column number within source
///\param jg row number within source
///\param w number of columns in global memory array
///\param h number of rows in global memory array
///\param p global memory array pitch in floats
///////////////////////////////////////////////////////////////////////////////
__forceinline__ __device__ void load_array(Ptr2D texSrc, float *smem, int ig, int jg, int w, int h, int p)
{
const int i = threadIdx.x + 2;
const int j = threadIdx.y + 2;
load_array_element(texSrc, smem, i, j, ig, jg, w, h, p);//load current pixel
__syncthreads();
if(threadIdx.y < 2)
{
//load bottom shadow elements
load_array_element(texSrc, smem, i, j-2, ig, jg-2, w, h, p);
if(threadIdx.x < 2)
{
//load bottom right shadow elements
load_array_element(texSrc, smem, i+PSOR_TILE_WIDTH, j-2, ig+PSOR_TILE_WIDTH, jg-2, w, h, p);
//load middle right shadow elements
load_array_element(texSrc, smem, i+PSOR_TILE_WIDTH, j, ig+PSOR_TILE_WIDTH, jg, w, h, p);
}
else if(threadIdx.x >= PSOR_TILE_WIDTH-2)
{
//load bottom left shadow elements
load_array_element(texSrc, smem, i-PSOR_TILE_WIDTH, j-2, ig-PSOR_TILE_WIDTH, jg-2, w, h, p);
//load middle left shadow elements
load_array_element(texSrc, smem, i-PSOR_TILE_WIDTH, j, ig-PSOR_TILE_WIDTH, jg, w, h, p);
}
}
else if(threadIdx.y >= PSOR_TILE_HEIGHT-2)
{
//load upper shadow elements
load_array_element(texSrc, smem, i, j+2, ig, jg+2, w, h, p);
if(threadIdx.x < 2)
{
//load upper right shadow elements
load_array_element(texSrc, smem, i+PSOR_TILE_WIDTH, j+2, ig+PSOR_TILE_WIDTH, jg+2, w, h, p);
//load middle right shadow elements
load_array_element(texSrc, smem, i+PSOR_TILE_WIDTH, j, ig+PSOR_TILE_WIDTH, jg, w, h, p);
}
else if(threadIdx.x >= PSOR_TILE_WIDTH-2)
{
//load upper left shadow elements
load_array_element(texSrc, smem, i-PSOR_TILE_WIDTH, j+2, ig-PSOR_TILE_WIDTH, jg+2, w, h, p);
//load middle left shadow elements
load_array_element(texSrc, smem, i-PSOR_TILE_WIDTH, j, ig-PSOR_TILE_WIDTH, jg, w, h, p);
}
}
else
{
//load middle shadow elements
if(threadIdx.x < 2)
{
//load middle right shadow elements
load_array_element(texSrc, smem, i+PSOR_TILE_WIDTH, j, ig+PSOR_TILE_WIDTH, jg, w, h, p);
}
else if(threadIdx.x >= PSOR_TILE_WIDTH-2)
{
//load middle left shadow elements
load_array_element(texSrc, smem, i-PSOR_TILE_WIDTH, j, ig-PSOR_TILE_WIDTH, jg, w, h, p);
}
}
__syncthreads();
}
///////////////////////////////////////////////////////////////////////////////
/// \brief computes matrix of linearised system for \c du, \c dv
/// Computed values reside in GPU memory. \n
/// Matrix computation is divided into two steps. This kernel performs first step\n
/// - compute smoothness term diffusivity between pixels - psi dash smooth
/// - compute robustness factor in the data term - psi dash data
/// \param diffusivity_x (in/out) diffusivity between pixels along x axis in smoothness term
/// \param diffusivity_y (in/out) diffusivity between pixels along y axis in smoothness term
/// \param denominator_u (in/out) precomputed part of expression for new du value in SOR iteration
/// \param denominator_v (in/out) precomputed part of expression for new dv value in SOR iteration
/// \param numerator_dudv (in/out) precomputed part of expression for new du and dv value in SOR iteration
/// \param numerator_u (in/out) precomputed part of expression for new du value in SOR iteration
/// \param numerator_v (in/out) precomputed part of expression for new dv value in SOR iteration
/// \param w (in) frame width
/// \param h (in) frame height
/// \param pitch (in) pitch in floats
/// \param alpha (in) alpha in Brox model (flow smoothness)
/// \param gamma (in) gamma in Brox model (edge importance)
///////////////////////////////////////////////////////////////////////////////
__global__ void prepare_sor_stage_1_tex(Ptr2D texU, Ptr2D texV, Ptr2D texDu, Ptr2D texDv, Ptr2D texI0, Ptr2D texI1, Ptr2D texIx, Ptr2D texIxx, Ptr2D texIx0, Ptr2D texIy, Ptr2D texIyy,
Ptr2D texIy0, Ptr2D texIxy, float *diffusivity_x, float *diffusivity_y, float *denominator_u, float *denominator_v, float *numerator_dudv, float *numerator_u, float *numerator_v,
int w, int h, int s, float alpha, float gamma)
{
__shared__ float u[PSOR_PITCH * PSOR_HEIGHT];
__shared__ float v[PSOR_PITCH * PSOR_HEIGHT];
__shared__ float du[PSOR_PITCH * PSOR_HEIGHT];
__shared__ float dv[PSOR_PITCH * PSOR_HEIGHT];
//position within tile
const int i = threadIdx.x;
const int j = threadIdx.y;
//position within smem arrays
const int ijs = (j+2) * PSOR_PITCH + i + 2;
//position within global memory
const int ig = blockIdx.x * blockDim.x + threadIdx.x;
const int jg = blockIdx.y * blockDim.y + threadIdx.y;
const int ijg = jg * s + ig;
//position within texture
float x = (float)ig + 0.5f;
float y = (float)jg + 0.5f;
//load u and v to smem
load_array(texU, u, ig, jg, w, h, s);
load_array(texV, v, ig, jg, w, h, s);
load_array(texDu, du, ig, jg, w, h, s);
load_array(texDv, dv, ig, jg, w, h, s);
//warped position
float wx = (x + u[ijs])/(float)w;
float wy = (y + v[ijs])/(float)h;
x /= (float)w;
y /= (float)h;
//compute image derivatives
const float Iz = texI1(wy, wx) - texI0(y,x);
const float Ix = texIx(wy, wx);
const float Ixz = Ix - texIx0(y, x);
const float Ixy = texIxy(wy, wx);
const float Ixx = texIxx(wy, wx);
const float Iy = texIy(wy, wx);
const float Iyz = Iy - texIy0(y, x);
const float Iyy = texIyy(wy, wx);
//compute data term
float q0, q1, q2;
q0 = Iz + Ix * du[ijs] + Iy * dv[ijs];
q1 = Ixz + Ixx * du[ijs] + Ixy * dv[ijs];
q2 = Iyz + Ixy * du[ijs] + Iyy * dv[ijs];
float data_term = 0.5f * rsqrtf(q0*q0 + gamma*(q1*q1 + q2*q2) + eps2);
//scale data term by 1/alpha
data_term /= alpha;
//compute smoothness term (diffusivity)
float sx, sy;
if(ig >= w || jg >= h) return;
diffusivity_along_x(&sx, ijs, u, v, du, dv);
diffusivity_along_y(&sy, ijs, u, v, du, dv);
if(ig == 0) sx = 0.0f;
if(jg == 0) sy = 0.0f;
numerator_dudv[ijg] = data_term * (Ix*Iy + gamma * Ixy*(Ixx + Iyy));
numerator_u[ijg] = data_term * (Ix*Iz + gamma * (Ixx*Ixz + Ixy*Iyz));
numerator_v[ijg] = data_term * (Iy*Iz + gamma * (Iyy*Iyz + Ixy*Ixz));
denominator_u[ijg] = data_term * (Ix*Ix + gamma * (Ixy*Ixy + Ixx*Ixx));
denominator_v[ijg] = data_term * (Iy*Iy + gamma * (Ixy*Ixy + Iyy*Iyy));
diffusivity_x[ijg] = sx;
diffusivity_y[ijg] = sy;
}
///////////////////////////////////////////////////////////////////////////////
///\brief computes matrix of linearised system for \c du, \c dv
///\param inv_denominator_u
///\param inv_denominator_v
///\param w
///\param h
///\param s
///////////////////////////////////////////////////////////////////////////////
__global__ void prepare_sor_stage_2(Ptr2D texDiffX, Ptr2D texDiffY, float *inv_denominator_u, float *inv_denominator_v, int w, int h, int s)
{
__shared__ float sx[(PSOR_TILE_WIDTH+1) * (PSOR_TILE_HEIGHT+1)];
__shared__ float sy[(PSOR_TILE_WIDTH+1) * (PSOR_TILE_HEIGHT+1)];
//position within tile
const int i = threadIdx.x;
const int j = threadIdx.y;
//position within smem arrays
const int ijs = j*(PSOR_TILE_WIDTH+1) + i;
//position within global memory
const int ig = blockIdx.x * blockDim.x + threadIdx.x;
const int jg = blockIdx.y * blockDim.y + threadIdx.y;
const int ijg = jg*s + ig;
int inside = ig < w && jg < h;
float denom_u;
float denom_v;
if(inside)
{
denom_u = inv_denominator_u[ijg];
denom_v = inv_denominator_v[ijg];
}
if(inside)
{
sx[ijs] = texDiffX(ijg);
sy[ijs] = texDiffY(ijg);
}
else
{
sx[ijs] = 0.0f;
sy[ijs] = 0.0f;
}
int up = ijs+PSOR_TILE_WIDTH+1;
if(j == PSOR_TILE_HEIGHT-1)
{
if(jg < h-1 && inside)
sy[up] = texDiffY(ijg + s);
else
sy[up] = 0.0f;
}
int right = ijs + 1;
if(threadIdx.x == PSOR_TILE_WIDTH-1)
{
if(ig < w-1 && inside)
sx[right] = texDiffX(ijg + 1);
else
sx[right] = 0.0f;
}
__syncthreads();
float diffusivity_sum;
diffusivity_sum = sx[ijs] + sx[ijs+1] + sy[ijs] + sy[ijs+PSOR_TILE_WIDTH+1];
if(inside)
{
denom_u += diffusivity_sum;
denom_v += diffusivity_sum;
inv_denominator_u[ijg] = 1.0f/denom_u;
inv_denominator_v[ijg] = 1.0f/denom_v;
}
}
/////////////////////////////////////////////////////////////////////////////////////////
// Red-Black SOR
/////////////////////////////////////////////////////////////////////////////////////////
template<int isBlack> __global__ void sor_pass(Ptr2D texU, Ptr2D texV, Ptr2D texDu, Ptr2D texDv, Ptr2D texDiffX, Ptr2D texDiffY, float *new_du, float *new_dv, const float *g_inv_denominator_u,
const float *g_inv_denominator_v, const float *g_numerator_u, const float *g_numerator_v, const float *g_numerator_dudv, float omega, int width, int height, int stride)
{
int i = blockIdx.x * blockDim.x + threadIdx.x;
int j = blockIdx.y * blockDim.y + threadIdx.y;
if(i >= width || j >= height)
return;
const int pos = j * stride + i;
const int pos_r = i < width - 1 ? pos + 1 : pos;
const int pos_u = j < height - 1 ? pos + stride : pos;
const int pos_d = j > 0 ? pos - stride : pos;
const int pos_l = i > 0 ? pos - 1 : pos;
//load smooth term
float s_up, s_left, s_right, s_down;
s_left = texDiffX(pos);
s_down = texDiffY(pos);
if(i < width-1)
s_right = texDiffX(pos_r);
else
s_right = 0.0f; //Neumann BC
if(j < height-1)
s_up = texDiffY(pos_u);
else
s_up = 0.0f; //Neumann BC
//load u, v and du, dv
float u_up, u_left, u_right, u_down, u;
float v_up, v_left, v_right, v_down, v;
float du_up, du_left, du_right, du_down, du;
float dv_up, dv_left, dv_right, dv_down, dv;
u_left = texU(pos_l);
u_right = texU(pos_r);
u_down = texU(pos_d);
u_up = texU(pos_u);
u = texU(pos);
v_left = texV(pos_l);
v_right = texV(pos_r);
v_down = texV(pos_d);
v = texV(pos);
v_up = texV(pos_u);
du = texDu(pos);
du_left = texDu(pos_l);
du_right = texDu(pos_r);
du_down = texDu(pos_d);
du_up = texDu(pos_u);
dv = texDv(pos);
dv_left = texDv(pos_l);
dv_right = texDv(pos_r);
dv_down = texDv(pos_d);
dv_up = texDv(pos_u);
float numerator_dudv = g_numerator_dudv[pos];
if((i+j)%2 == isBlack)
{
// update du
float numerator_u = (s_left*(u_left + du_left) + s_up*(u_up + du_up) + s_right*(u_right + du_right) + s_down*(u_down + du_down) -
u * (s_left + s_right + s_up + s_down) - g_numerator_u[pos] - numerator_dudv*dv);
du = (1.0f - omega) * du + omega * g_inv_denominator_u[pos] * numerator_u;
// update dv
float numerator_v = (s_left*(v_left + dv_left) + s_up*(v_up + dv_up) + s_right*(v_right + dv_right) + s_down*(v_down + dv_down) -
v * (s_left + s_right + s_up + s_down) - g_numerator_v[pos] - numerator_dudv*du);
dv = (1.0f - omega) * dv + omega * g_inv_denominator_v[pos] * numerator_v;
}
new_du[pos] = du;
new_dv[pos] = dv;
}
///////////////////////////////////////////////////////////////////////////////
// utility functions
///////////////////////////////////////////////////////////////////////////////
namespace
{
struct ImagePyramid
{
std::vector<FloatVector*> img0;
std::vector<FloatVector*> img1;
std::vector<Ncv32u> w;
std::vector<Ncv32u> h;
explicit ImagePyramid(int outer_iterations)
{
img0.reserve(outer_iterations);
img1.reserve(outer_iterations);
w.reserve(outer_iterations);
h.reserve(outer_iterations);
}
~ImagePyramid()
{
w.clear();
h.clear();
for (int i = static_cast<int>(img0.size()) - 1; i >= 0; --i)
{
delete img1[i];
delete img0[i];
}
img0.clear();
img1.clear();
}
};
}
/////////////////////////////////////////////////////////////////////////////////////////
// MAIN FUNCTION
/////////////////////////////////////////////////////////////////////////////////////////
NCVStatus NCVBroxOpticalFlow(const NCVBroxOpticalFlowDescriptor desc,
INCVMemAllocator &gpu_mem_allocator,
const NCVMatrix<Ncv32f> &frame0,
const NCVMatrix<Ncv32f> &frame1,
NCVMatrix<Ncv32f> &uOut,
NCVMatrix<Ncv32f> &vOut,
cudaStream_t stream)
{
ncvAssertPrintReturn(desc.alpha > 0.0f , "Invalid alpha" , NCV_INCONSISTENT_INPUT);
ncvAssertPrintReturn(desc.gamma >= 0.0f , "Invalid gamma" , NCV_INCONSISTENT_INPUT);
ncvAssertPrintReturn(desc.number_of_inner_iterations > 0 , "Invalid number of inner iterations" , NCV_INCONSISTENT_INPUT);
ncvAssertPrintReturn(desc.number_of_outer_iterations > 0 , "Invalid number of outer iterations" , NCV_INCONSISTENT_INPUT);
ncvAssertPrintReturn(desc.number_of_solver_iterations > 0, "Invalid number of solver iterations", NCV_INCONSISTENT_INPUT);
const Ncv32u kSourceWidth = frame0.width();
const Ncv32u kSourceHeight = frame0.height();
ncvAssertPrintReturn(frame1.width() == kSourceWidth && frame1.height() == kSourceHeight, "Frame dims do not match", NCV_INCONSISTENT_INPUT);
ncvAssertReturn(uOut.width() == kSourceWidth && vOut.width() == kSourceWidth &&
uOut.height() == kSourceHeight && vOut.height() == kSourceHeight, NCV_INCONSISTENT_INPUT);
ncvAssertReturn(gpu_mem_allocator.isInitialized(), NCV_ALLOCATOR_NOT_INITIALIZED);
bool kSkipProcessing = gpu_mem_allocator.isCounting();
int cuda_device;
ncvAssertCUDAReturn(cudaGetDevice(&cuda_device), NCV_CUDA_ERROR);
cudaDeviceProp device_props;
ncvAssertCUDAReturn(cudaGetDeviceProperties(&device_props, cuda_device), NCV_CUDA_ERROR);
Ncv32u alignmentValue = gpu_mem_allocator.alignment ();
const Ncv32u kStrideAlignmentFloat = alignmentValue / sizeof(float);
const Ncv32u kSourcePitch = alignUp(kSourceWidth, kStrideAlignmentFloat) * sizeof(float);
const Ncv32f scale_factor = desc.scale_factor;
const Ncv32f alpha = desc.alpha;
const Ncv32f gamma = desc.gamma;
const Ncv32u kSizeInPixelsAligned = alignUp(kSourceWidth, kStrideAlignmentFloat)*kSourceHeight;
#if defined SAFE_VECTOR_DECL
#undef SAFE_VECTOR_DECL
#endif
#define SAFE_VECTOR_DECL(name, allocator, size) \
FloatVector name((allocator), (size)); \
ncvAssertReturn(name.isMemAllocated(), NCV_ALLOCATOR_BAD_ALLOC);
// matrix elements
SAFE_VECTOR_DECL(diffusivity_x, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(diffusivity_y, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(denom_u, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(denom_v, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(num_dudv, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(num_u, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(num_v, gpu_mem_allocator, kSizeInPixelsAligned);
// flow components
SAFE_VECTOR_DECL(u, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(v, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(u_new, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(v_new, gpu_mem_allocator, kSizeInPixelsAligned);
// flow increments
SAFE_VECTOR_DECL(du, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(dv, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(du_new, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(dv_new, gpu_mem_allocator, kSizeInPixelsAligned);
// temporary storage
SAFE_VECTOR_DECL(device_buffer, gpu_mem_allocator,
alignUp(kSourceWidth, kStrideAlignmentFloat) * alignUp(kSourceHeight, kStrideAlignmentFloat));
// image derivatives
SAFE_VECTOR_DECL(Ix, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(Ixx, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(Ix0, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(Iy, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(Iyy, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(Iy0, gpu_mem_allocator, kSizeInPixelsAligned);
SAFE_VECTOR_DECL(Ixy, gpu_mem_allocator, kSizeInPixelsAligned);
// spatial derivative filter size
const int kDFilterSize = 5;
SAFE_VECTOR_DECL(derivativeFilter, gpu_mem_allocator, kDFilterSize);
if (!kSkipProcessing)
{
const float derivativeFilterHost[kDFilterSize] = {1.0f, -8.0f, 0.0f, 8.0f, -1.0f};
ncvAssertCUDAReturn(cudaMemcpy(derivativeFilter.ptr(), derivativeFilterHost, sizeof(float) * kDFilterSize,
cudaMemcpyHostToDevice), NCV_CUDA_ERROR);
}
//prepare image pyramid
ImagePyramid pyr(desc.number_of_outer_iterations);
float scale = 1.0f;
//cuda arrays for frames
std::unique_ptr<FloatVector> pI0(new FloatVector(gpu_mem_allocator, kSizeInPixelsAligned));
ncvAssertReturn(pI0->isMemAllocated(), NCV_ALLOCATOR_BAD_ALLOC);
std::unique_ptr<FloatVector> pI1(new FloatVector(gpu_mem_allocator, kSizeInPixelsAligned));
ncvAssertReturn(pI1->isMemAllocated(), NCV_ALLOCATOR_BAD_ALLOC);
if (!kSkipProcessing)
{
//copy frame data to device
size_t dst_width_in_bytes = alignUp(kSourceWidth, kStrideAlignmentFloat) * sizeof(float);
size_t src_width_in_bytes = kSourceWidth * sizeof(float);
size_t src_pitch_in_bytes = frame0.pitch();
ncvAssertCUDAReturn( cudaMemcpy2DAsync(pI0->ptr(), dst_width_in_bytes, frame0.ptr(),
src_pitch_in_bytes, src_width_in_bytes, kSourceHeight, cudaMemcpyDeviceToDevice, stream), NCV_CUDA_ERROR );
ncvAssertCUDAReturn( cudaMemcpy2DAsync(pI1->ptr(), dst_width_in_bytes, frame1.ptr(),
src_pitch_in_bytes, src_width_in_bytes, kSourceHeight, cudaMemcpyDeviceToDevice, stream), NCV_CUDA_ERROR );
}
FloatVector* I0 = pI0.release();
FloatVector* I1 = pI1.release();
//prepare pyramid
pyr.img0.push_back(I0);
pyr.img1.push_back(I1);
pyr.w.push_back(kSourceWidth);
pyr.h.push_back(kSourceHeight);
scale *= scale_factor;
Ncv32u prev_level_width = kSourceWidth;
Ncv32u prev_level_height = kSourceHeight;
while((prev_level_width > 15) && (prev_level_height > 15) && (static_cast<Ncv32u>(pyr.img0.size()) < desc.number_of_outer_iterations))
{
//current resolution
Ncv32u level_width = static_cast<Ncv32u>(ceilf(kSourceWidth * scale));
Ncv32u level_height = static_cast<Ncv32u>(ceilf(kSourceHeight * scale));
Ncv32u level_width_aligned = alignUp(level_width, kStrideAlignmentFloat);
Ncv32u buffer_size = alignUp(level_width, kStrideAlignmentFloat) * level_height; // buffer size in floats
Ncv32u prev_level_pitch = alignUp(prev_level_width, kStrideAlignmentFloat) * sizeof(float);
std::unique_ptr<FloatVector> level_frame0(new FloatVector(gpu_mem_allocator, buffer_size));
ncvAssertReturn(level_frame0->isMemAllocated(), NCV_ALLOCATOR_BAD_ALLOC);
std::unique_ptr<FloatVector> level_frame1(new FloatVector(gpu_mem_allocator, buffer_size));
ncvAssertReturn(level_frame1->isMemAllocated(), NCV_ALLOCATOR_BAD_ALLOC);
if (!kSkipProcessing)
{
ncvAssertCUDAReturn(cudaStreamSynchronize(stream), NCV_CUDA_ERROR);
NcvSize32u srcSize (prev_level_width, prev_level_height);
NcvSize32u dstSize (level_width, level_height);
NcvRect32u srcROI (0, 0, prev_level_width, prev_level_height);
NcvRect32u dstROI (0, 0, level_width, level_height);
// frame 0
ncvAssertReturnNcvStat( nppiStResize_32f_C1R (I0->ptr(), srcSize, prev_level_pitch, srcROI,
level_frame0->ptr(), dstSize, level_width_aligned * sizeof (float), dstROI, scale_factor, scale_factor, nppStSupersample) );
// frame 1
ncvAssertReturnNcvStat( nppiStResize_32f_C1R (I1->ptr(), srcSize, prev_level_pitch, srcROI,
level_frame1->ptr(), dstSize, level_width_aligned * sizeof (float), dstROI, scale_factor, scale_factor, nppStSupersample) );
}
I0 = level_frame0.release();
I1 = level_frame1.release();
//store pointers
pyr.img0.push_back(I0);
pyr.img1.push_back(I1);
pyr.w.push_back(level_width);
pyr.h.push_back(level_height);
scale *= scale_factor;
prev_level_width = level_width;
prev_level_height = level_height;
}
if (!kSkipProcessing)
{
//initial values for flow is 0
ncvAssertCUDAReturn(cudaMemsetAsync(u.ptr(), 0, kSizeInPixelsAligned * sizeof(float), stream), NCV_CUDA_ERROR);
ncvAssertCUDAReturn(cudaMemsetAsync(v.ptr(), 0, kSizeInPixelsAligned * sizeof(float), stream), NCV_CUDA_ERROR);
//select images with lowest resolution
ncvAssertCUDAReturn(cudaStreamSynchronize(stream), NCV_CUDA_ERROR);
FloatVector* ptrU = &u;
FloatVector* ptrV = &v;
FloatVector* ptrUNew = &u_new;
FloatVector* ptrVNew = &v_new;
std::vector<FloatVector*>::const_reverse_iterator img0Iter = pyr.img0.rbegin();
std::vector<FloatVector*>::const_reverse_iterator img1Iter = pyr.img1.rbegin();
//outer loop
//warping fixed point iteration
while(!pyr.w.empty())
{
//current grid dimensions
const Ncv32u kLevelWidth = pyr.w.back();
const Ncv32u kLevelHeight = pyr.h.back();
const Ncv32u kLevelStride = alignUp(kLevelWidth, kStrideAlignmentFloat);
//size of current image in bytes
const int kLevelSizeInBytes = kLevelStride * kLevelHeight * sizeof(float);
//number of points at current resolution
const int kLevelSizeInPixels = kLevelStride * kLevelHeight;
//initial guess for du and dv
ncvAssertCUDAReturn(cudaMemsetAsync(du.ptr(), 0, kLevelSizeInBytes, stream), NCV_CUDA_ERROR);
ncvAssertCUDAReturn(cudaMemsetAsync(dv.ptr(), 0, kLevelSizeInBytes, stream), NCV_CUDA_ERROR);
I0 = *img0Iter;
I1 = *img1Iter;
++img0Iter;
++img1Iter;
Texture texI0(kLevelHeight, kLevelWidth, I0->ptr(), kLevelStride * sizeof(float), true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texI1(kLevelHeight, kLevelWidth, I1->ptr(), kLevelStride * sizeof(float), true, cudaFilterModeLinear, cudaAddressModeMirror);
//compute derivatives
dim3 dBlocks(iDivUp(kLevelWidth, 32), iDivUp(kLevelHeight, 6));
//dim3 dThreads(32, 6);
const int kPitchTex = kLevelStride * sizeof(float);
NcvSize32u srcSize(kLevelWidth, kLevelHeight);
Ncv32u nSrcStep = kLevelStride * sizeof(float);
NcvRect32u oROI(0, 0, kLevelWidth, kLevelHeight);
// Ix0
ncvAssertReturnNcvStat( nppiStFilterRowBorder_32f_C1R (I0->ptr(), srcSize, nSrcStep, Ix0.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
// Iy0
ncvAssertReturnNcvStat( nppiStFilterColumnBorder_32f_C1R (I0->ptr(), srcSize, nSrcStep, Iy0.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
// Ix
ncvAssertReturnNcvStat( nppiStFilterRowBorder_32f_C1R (I1->ptr(), srcSize, nSrcStep, Ix.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
// Iy
ncvAssertReturnNcvStat( nppiStFilterColumnBorder_32f_C1R (I1->ptr(), srcSize, nSrcStep, Iy.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
// Ixx
ncvAssertReturnNcvStat( nppiStFilterRowBorder_32f_C1R (Ix.ptr(), srcSize, nSrcStep, Ixx.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
// Iyy
ncvAssertReturnNcvStat( nppiStFilterColumnBorder_32f_C1R (Iy.ptr(), srcSize, nSrcStep, Iyy.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
// Ixy
ncvAssertReturnNcvStat( nppiStFilterRowBorder_32f_C1R (Iy.ptr(), srcSize, nSrcStep, Ixy.ptr(), srcSize, nSrcStep, oROI,
nppStBorderMirror, derivativeFilter.ptr(), kDFilterSize, kDFilterSize/2, 1.0f/12.0f) );
Texture texIx(kLevelHeight, kLevelWidth, Ix.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texIxx(kLevelHeight, kLevelWidth, Ixx.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texIx0(kLevelHeight, kLevelWidth, Ix0.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texIy(kLevelHeight, kLevelWidth, Iy.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texIyy(kLevelHeight, kLevelWidth, Iyy.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texIy0(kLevelHeight, kLevelWidth, Iy0.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texIxy(kLevelHeight, kLevelWidth, Ixy.ptr(), kPitchTex, true, cudaFilterModeLinear, cudaAddressModeMirror);
Texture texDiffX(kLevelSizeInBytes, diffusivity_x.ptr());
Texture texDiffY(kLevelSizeInBytes, diffusivity_y.ptr());
// flow
Texture texU(kLevelSizeInBytes, ptrU->ptr());
Texture texV(kLevelSizeInBytes, ptrV->ptr());
// flow increments
Texture texDu(kLevelSizeInBytes, du.ptr());
Texture texDv(kLevelSizeInBytes, dv.ptr());
Texture texDuNew(kLevelSizeInBytes, du_new.ptr());
Texture texDvNew(kLevelSizeInBytes, dv_new.ptr());
dim3 psor_blocks(iDivUp(kLevelWidth, PSOR_TILE_WIDTH), iDivUp(kLevelHeight, PSOR_TILE_HEIGHT));
dim3 psor_threads(PSOR_TILE_WIDTH, PSOR_TILE_HEIGHT);
dim3 sor_blocks(iDivUp(kLevelWidth, SOR_TILE_WIDTH), iDivUp(kLevelHeight, SOR_TILE_HEIGHT));
dim3 sor_threads(SOR_TILE_WIDTH, SOR_TILE_HEIGHT);
// inner loop
// lagged nonlinearity fixed point iteration
ncvAssertCUDAReturn(cudaStreamSynchronize(stream), NCV_CUDA_ERROR);
for (Ncv32u current_inner_iteration = 0; current_inner_iteration < desc.number_of_inner_iterations; ++current_inner_iteration)
{
//compute coefficients
prepare_sor_stage_1_tex<<<psor_blocks, psor_threads, 0, stream>>> (texU, texV, texDu, texDv, texI0, texI1, texIx, texIxx, texIx0, texIy, texIyy, texIy0, texIxy,
diffusivity_x.ptr(), diffusivity_y.ptr(), denom_u.ptr(), denom_v.ptr(), num_dudv.ptr(), num_u.ptr(), num_v.ptr(), kLevelWidth, kLevelHeight, kLevelStride, alpha, gamma);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
prepare_sor_stage_2<<<psor_blocks, psor_threads, 0, stream>>>(texDiffX, texDiffY, denom_u.ptr(), denom_v.ptr(), kLevelWidth, kLevelHeight, kLevelStride);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
//solve linear system
for (Ncv32u solver_iteration = 0; solver_iteration < desc.number_of_solver_iterations; ++solver_iteration)
{
float omega = 1.99f;
sor_pass<0><<<sor_blocks, sor_threads, 0, stream>>>(texU, texV, texDu, texDv, texDiffX, texDiffY, du_new.ptr(), dv_new.ptr(), denom_u.ptr(), denom_v.ptr(),
num_u.ptr(), num_v.ptr(), num_dudv.ptr(), omega, kLevelWidth, kLevelHeight, kLevelStride);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
sor_pass<1><<<sor_blocks, sor_threads, 0, stream>>>(texU, texV, texDuNew, texDvNew, texDiffX, texDiffY, du.ptr(), dv.ptr(), denom_u.ptr(), denom_v.ptr(), num_u.ptr(),
num_v.ptr(),num_dudv.ptr(), omega, kLevelWidth, kLevelHeight, kLevelStride);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
}//end of solver loop
}// end of inner loop
//update u and v
add(ptrU->ptr(), du.ptr(), kLevelSizeInPixels, stream);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
add(ptrV->ptr(), dv.ptr(), kLevelSizeInPixels, stream);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
//prolongate using texture
pyr.w.pop_back();
pyr.h.pop_back();
if (!pyr.w.empty())
{
//compute new image size
Ncv32u nw = pyr.w.back();
Ncv32u nh = pyr.h.back();
Ncv32u ns = alignUp(nw, kStrideAlignmentFloat);
dim3 p_blocks(iDivUp(nw, 32), iDivUp(nh, 8));
//dim3 p_threads(32, 8);
NcvSize32u inner_srcSize (kLevelWidth, kLevelHeight);
NcvSize32u dstSize (nw, nh);
NcvRect32u srcROI (0, 0, kLevelWidth, kLevelHeight);
NcvRect32u dstROI (0, 0, nw, nh);
ncvAssertReturnNcvStat( nppiStResize_32f_C1R (ptrU->ptr(), inner_srcSize, kLevelStride * sizeof (float), srcROI,
ptrUNew->ptr(), dstSize, ns * sizeof (float), dstROI, 1.0f/scale_factor, 1.0f/scale_factor, nppStBicubic) );
ScaleVector(ptrUNew->ptr(), ptrUNew->ptr(), 1.0f/scale_factor, ns * nh, stream);
ncvAssertCUDALastErrorReturn(NCV_CUDA_ERROR);
ncvAssertReturnNcvStat( nppiStResize_32f_C1R (ptrV->ptr(), inner_srcSize, kLevelStride * sizeof (float), srcROI,
ptrVNew->ptr(), dstSize, ns * sizeof (float), dstROI, 1.0f/scale_factor, 1.0f/scale_factor, nppStBicubic) );
ScaleVector(ptrVNew->ptr(), ptrVNew->ptr(), 1.0f/scale_factor, ns * nh, stream);
ncvAssertCUDALastErrorReturn((int)NCV_CUDA_ERROR);
cv::cuda::device::swap<FloatVector*>(ptrU, ptrUNew);
cv::cuda::device::swap<FloatVector*>(ptrV, ptrVNew);
}
scale /= scale_factor;
}
// end of warping iterations
ncvAssertCUDAReturn(cudaStreamSynchronize(stream), (int)NCV_CUDA_ERROR);
ncvAssertCUDAReturn( cudaMemcpy2DAsync
(uOut.ptr(), uOut.pitch(), ptrU->ptr(),
kSourcePitch, kSourceWidth*sizeof(float), kSourceHeight, cudaMemcpyDeviceToDevice, stream), (int)NCV_CUDA_ERROR );
ncvAssertCUDAReturn( cudaMemcpy2DAsync
(vOut.ptr(), vOut.pitch(), ptrV->ptr(),
kSourcePitch, kSourceWidth*sizeof(float), kSourceHeight, cudaMemcpyDeviceToDevice, stream), (int)NCV_CUDA_ERROR );
ncvAssertCUDAReturn(cudaStreamSynchronize(stream), (int)NCV_CUDA_ERROR);
}
return NCV_SUCCESS;
}
modules/cudalegacy/src/cuda/NCVColorConversion.hpp
+100
View File
@@ -0,0 +1,100 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
// this file does not contain any used code.
#ifndef _ncv_color_conversion_hpp_
#define _ncv_color_conversion_hpp_
#include "NCVPixelOperations.hpp"
#if 0
enum NCVColorSpace
{
NCVColorSpaceGray,
NCVColorSpaceRGBA,
};
template<NCVColorSpace CSin, NCVColorSpace CSout, typename Tin, typename Tout> struct __pixColorConv {
static void _pixColorConv(const Tin &pixIn, Tout &pixOut);
};
template<typename Tin, typename Tout> struct __pixColorConv<NCVColorSpaceRGBA, NCVColorSpaceGray, Tin, Tout> {
static void _pixColorConv(const Tin &pixIn, Tout &pixOut)
{
Ncv32f luma = 0.299f * pixIn.x + 0.587f * pixIn.y + 0.114f * pixIn.z;
_TDemoteClampNN(luma, pixOut.x);
}};
template<typename Tin, typename Tout> struct __pixColorConv<NCVColorSpaceGray, NCVColorSpaceRGBA, Tin, Tout> {
static void _pixColorConv(const Tin &pixIn, Tout &pixOut)
{
_TDemoteClampNN(pixIn.x, pixOut.x);
_TDemoteClampNN(pixIn.x, pixOut.y);
_TDemoteClampNN(pixIn.x, pixOut.z);
pixOut.w = 0;
}};
template<NCVColorSpace CSin, NCVColorSpace CSout, typename Tin, typename Tout>
static NCVStatus _ncvColorConv_host(const NCVMatrix<Tin> &h_imgIn,
const NCVMatrix<Tout> &h_imgOut)
{
ncvAssertReturn(h_imgIn.size() == h_imgOut.size(), NCV_DIMENSIONS_INVALID);
ncvAssertReturn(h_imgIn.memType() == h_imgOut.memType() &&
(h_imgIn.memType() == NCVMemoryTypeHostPinned || h_imgIn.memType() == NCVMemoryTypeNone), NCV_MEM_RESIDENCE_ERROR);
NCV_SET_SKIP_COND(h_imgIn.memType() == NCVMemoryTypeNone);
NCV_SKIP_COND_BEGIN
for (Ncv32u i=0; i<h_imgIn.height(); i++)
{
for (Ncv32u j=0; j<h_imgIn.width(); j++)
{
__pixColorConv<CSin, CSout, Tin, Tout>::_pixColorConv(h_imgIn.at(j,i), h_imgOut.at(j,i));
}
}
NCV_SKIP_COND_END
return NCV_SUCCESS;
}
#endif
#endif //_ncv_color_conversion_hpp_
modules/cudalegacy/src/cuda/NCVHaarObjectDetection.cu
+2291
View File
File diff suppressed because it is too large Load Diff
modules/cudalegacy/src/cuda/NCVPixelOperations.hpp
+354
View File
@@ -0,0 +1,354 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncv_pixel_operations_hpp_
#define _ncv_pixel_operations_hpp_
#include <limits.h>
#include <float.h>
#include "opencv2/cudalegacy/NCV.hpp"
#include "opencv2/core/cuda/cuda_compat.hpp"
using cv::cuda::device::compat::double4Compat;
using cv::cuda::device::compat::make_double4_compat;
template<typename TBase> inline __host__ __device__ TBase _pixMaxVal();
template<> inline __host__ __device__ Ncv8u _pixMaxVal<Ncv8u>() {return UCHAR_MAX;}
template<> inline __host__ __device__ Ncv16u _pixMaxVal<Ncv16u>() {return USHRT_MAX;}
template<> inline __host__ __device__ Ncv32u _pixMaxVal<Ncv32u>() {return UINT_MAX;}
template<> inline __host__ __device__ Ncv8s _pixMaxVal<Ncv8s>() {return SCHAR_MAX;}
template<> inline __host__ __device__ Ncv16s _pixMaxVal<Ncv16s>() {return SHRT_MAX;}
template<> inline __host__ __device__ Ncv32s _pixMaxVal<Ncv32s>() {return INT_MAX;}
template<> inline __host__ __device__ Ncv32f _pixMaxVal<Ncv32f>() {return FLT_MAX;}
template<> inline __host__ __device__ Ncv64f _pixMaxVal<Ncv64f>() {return DBL_MAX;}
template<typename TBase> inline __host__ __device__ TBase _pixMinVal();
template<> inline __host__ __device__ Ncv8u _pixMinVal<Ncv8u>() {return 0;}
template<> inline __host__ __device__ Ncv16u _pixMinVal<Ncv16u>() {return 0;}
template<> inline __host__ __device__ Ncv32u _pixMinVal<Ncv32u>() {return 0;}
template<> inline __host__ __device__ Ncv8s _pixMinVal<Ncv8s>() {return SCHAR_MIN;}
template<> inline __host__ __device__ Ncv16s _pixMinVal<Ncv16s>() {return SHRT_MIN;}
template<> inline __host__ __device__ Ncv32s _pixMinVal<Ncv32s>() {return INT_MIN;}
template<> inline __host__ __device__ Ncv32f _pixMinVal<Ncv32f>() {return FLT_MIN;}
template<> inline __host__ __device__ Ncv64f _pixMinVal<Ncv64f>() {return DBL_MIN;}
template<typename Tvec> struct TConvVec2Base;
template<> struct TConvVec2Base<uchar1> {typedef Ncv8u TBase;};
template<> struct TConvVec2Base<uchar3> {typedef Ncv8u TBase;};
template<> struct TConvVec2Base<uchar4> {typedef Ncv8u TBase;};
template<> struct TConvVec2Base<ushort1> {typedef Ncv16u TBase;};
template<> struct TConvVec2Base<ushort3> {typedef Ncv16u TBase;};
template<> struct TConvVec2Base<ushort4> {typedef Ncv16u TBase;};
template<> struct TConvVec2Base<uint1> {typedef Ncv32u TBase;};
template<> struct TConvVec2Base<uint3> {typedef Ncv32u TBase;};
template<> struct TConvVec2Base<uint4> {typedef Ncv32u TBase;};
template<> struct TConvVec2Base<float1> {typedef Ncv32f TBase;};
template<> struct TConvVec2Base<float3> {typedef Ncv32f TBase;};
template<> struct TConvVec2Base<float4> {typedef Ncv32f TBase;};
template<> struct TConvVec2Base<double1> {typedef Ncv64f TBase;};
template<> struct TConvVec2Base<double3> {typedef Ncv64f TBase;};
template<> struct TConvVec2Base<double4> {typedef Ncv64f TBase;};
#define NC(T) (sizeof(T) / sizeof(typename TConvVec2Base<T>::TBase))
template<typename TBase, Ncv32u NC> struct TConvBase2Vec;
template<> struct TConvBase2Vec<Ncv8u, 1> {typedef uchar1 TVec;};
template<> struct TConvBase2Vec<Ncv8u, 3> {typedef uchar3 TVec;};
template<> struct TConvBase2Vec<Ncv8u, 4> {typedef uchar4 TVec;};
template<> struct TConvBase2Vec<Ncv16u, 1> {typedef ushort1 TVec;};
template<> struct TConvBase2Vec<Ncv16u, 3> {typedef ushort3 TVec;};
template<> struct TConvBase2Vec<Ncv16u, 4> {typedef ushort4 TVec;};
template<> struct TConvBase2Vec<Ncv32u, 1> {typedef uint1 TVec;};
template<> struct TConvBase2Vec<Ncv32u, 3> {typedef uint3 TVec;};
template<> struct TConvBase2Vec<Ncv32u, 4> {typedef uint4 TVec;};
template<> struct TConvBase2Vec<Ncv32f, 1> {typedef float1 TVec;};
template<> struct TConvBase2Vec<Ncv32f, 3> {typedef float3 TVec;};
template<> struct TConvBase2Vec<Ncv32f, 4> {typedef float4 TVec;};
template<> struct TConvBase2Vec<Ncv64f, 1> {typedef double1 TVec;};
template<> struct TConvBase2Vec<Ncv64f, 3> {typedef double3 TVec;};
template<> struct TConvBase2Vec<Ncv64f, 4> {typedef double4Compat TVec;};
//TODO: consider using CUDA intrinsics to avoid branching
template<typename Tin> inline __host__ __device__ void _TDemoteClampZ(Tin &a, Ncv8u &out) {out = (Ncv8u)CLAMP_0_255(a);}
template<typename Tin> inline __host__ __device__ void _TDemoteClampZ(Tin &a, Ncv16u &out) {out = (Ncv16u)CLAMP(a, 0, USHRT_MAX);}
template<typename Tin> inline __host__ __device__ void _TDemoteClampZ(Tin &a, Ncv32u &out) {out = (Ncv32u)CLAMP(a, 0, UINT_MAX);}
template<typename Tin> inline __host__ __device__ void _TDemoteClampZ(Tin &a, Ncv32f &out) {out = (Ncv32f)a;}
//TODO: consider using CUDA intrinsics to avoid branching
template<typename Tin> inline __host__ __device__ void _TDemoteClampNN(Tin &a, Ncv8u &out) {out = (Ncv8u)CLAMP_0_255(a+0.5f);}
template<typename Tin> inline __host__ __device__ void _TDemoteClampNN(Tin &a, Ncv16u &out) {out = (Ncv16u)CLAMP(a+0.5f, 0, USHRT_MAX);}
template<typename Tin> inline __host__ __device__ void _TDemoteClampNN(Tin &a, Ncv32u &out) {out = (Ncv32u)CLAMP(a+0.5f, 0, UINT_MAX);}
template<typename Tin> inline __host__ __device__ void _TDemoteClampNN(Tin &a, Ncv32f &out) {out = (Ncv32f)a;}
template<typename Tout> inline __host__ __device__ Tout _pixMakeZero();
template<> inline __host__ __device__ uchar1 _pixMakeZero<uchar1>() {return make_uchar1(0);}
template<> inline __host__ __device__ uchar3 _pixMakeZero<uchar3>() {return make_uchar3(0,0,0);}
template<> inline __host__ __device__ uchar4 _pixMakeZero<uchar4>() {return make_uchar4(0,0,0,0);}
template<> inline __host__ __device__ ushort1 _pixMakeZero<ushort1>() {return make_ushort1(0);}
template<> inline __host__ __device__ ushort3 _pixMakeZero<ushort3>() {return make_ushort3(0,0,0);}
template<> inline __host__ __device__ ushort4 _pixMakeZero<ushort4>() {return make_ushort4(0,0,0,0);}
template<> inline __host__ __device__ uint1 _pixMakeZero<uint1>() {return make_uint1(0);}
template<> inline __host__ __device__ uint3 _pixMakeZero<uint3>() {return make_uint3(0,0,0);}
template<> inline __host__ __device__ uint4 _pixMakeZero<uint4>() {return make_uint4(0,0,0,0);}
template<> inline __host__ __device__ float1 _pixMakeZero<float1>() {return make_float1(0.f);}
template<> inline __host__ __device__ float3 _pixMakeZero<float3>() {return make_float3(0.f,0.f,0.f);}
template<> inline __host__ __device__ float4 _pixMakeZero<float4>() {return make_float4(0.f,0.f,0.f,0.f);}
template<> inline __host__ __device__ double1 _pixMakeZero<double1>() {return make_double1(0.);}
template<> inline __host__ __device__ double3 _pixMakeZero<double3>() {return make_double3(0.,0.,0.);}
template<> inline __host__ __device__ double4Compat _pixMakeZero<double4Compat>() {return make_double4_compat(0.,0.,0.,0.);}
static inline __host__ __device__ uchar1 _pixMake(Ncv8u x) {return make_uchar1(x);}
static inline __host__ __device__ uchar3 _pixMake(Ncv8u x, Ncv8u y, Ncv8u z) {return make_uchar3(x,y,z);}
static inline __host__ __device__ uchar4 _pixMake(Ncv8u x, Ncv8u y, Ncv8u z, Ncv8u w) {return make_uchar4(x,y,z,w);}
static inline __host__ __device__ ushort1 _pixMake(Ncv16u x) {return make_ushort1(x);}
static inline __host__ __device__ ushort3 _pixMake(Ncv16u x, Ncv16u y, Ncv16u z) {return make_ushort3(x,y,z);}
static inline __host__ __device__ ushort4 _pixMake(Ncv16u x, Ncv16u y, Ncv16u z, Ncv16u w) {return make_ushort4(x,y,z,w);}
static inline __host__ __device__ uint1 _pixMake(Ncv32u x) {return make_uint1(x);}
static inline __host__ __device__ uint3 _pixMake(Ncv32u x, Ncv32u y, Ncv32u z) {return make_uint3(x,y,z);}
static inline __host__ __device__ uint4 _pixMake(Ncv32u x, Ncv32u y, Ncv32u z, Ncv32u w) {return make_uint4(x,y,z,w);}
static inline __host__ __device__ float1 _pixMake(Ncv32f x) {return make_float1(x);}
static inline __host__ __device__ float3 _pixMake(Ncv32f x, Ncv32f y, Ncv32f z) {return make_float3(x,y,z);}
static inline __host__ __device__ float4 _pixMake(Ncv32f x, Ncv32f y, Ncv32f z, Ncv32f w) {return make_float4(x,y,z,w);}
static inline __host__ __device__ double1 _pixMake(Ncv64f x) {return make_double1(x);}
static inline __host__ __device__ double3 _pixMake(Ncv64f x, Ncv64f y, Ncv64f z) {return make_double3(x,y,z);}
static inline __host__ __device__ double4Compat _pixMake(Ncv64f x, Ncv64f y, Ncv64f z, Ncv64f w) {return make_double4_compat(x,y,z,w);}
template<typename Tin, typename Tout, Ncv32u CN> struct __pixDemoteClampZ_CN {static __host__ __device__ Tout _pixDemoteClampZ_CN(Tin &pix);};
template<typename Tin, typename Tout> struct __pixDemoteClampZ_CN<Tin, Tout, 1> {
static __host__ __device__ Tout _pixDemoteClampZ_CN(Tin &pix)
{
Tout out;
_TDemoteClampZ(pix.x, out.x);
return out;
}};
template<typename Tin, typename Tout> struct __pixDemoteClampZ_CN<Tin, Tout, 3> {
static __host__ __device__ Tout _pixDemoteClampZ_CN(Tin &pix)
{
Tout out;
_TDemoteClampZ(pix.x, out.x);
_TDemoteClampZ(pix.y, out.y);
_TDemoteClampZ(pix.z, out.z);
return out;
}};
template<typename Tin, typename Tout> struct __pixDemoteClampZ_CN<Tin, Tout, 4> {
static __host__ __device__ Tout _pixDemoteClampZ_CN(Tin &pix)
{
Tout out;
_TDemoteClampZ(pix.x, out.x);
_TDemoteClampZ(pix.y, out.y);
_TDemoteClampZ(pix.z, out.z);
_TDemoteClampZ(pix.w, out.w);
return out;
}};
template<typename Tin, typename Tout> inline __host__ __device__ Tout _pixDemoteClampZ(Tin &pix)
{
return __pixDemoteClampZ_CN<Tin, Tout, NC(Tin)>::_pixDemoteClampZ_CN(pix);
}
template<typename Tin, typename Tout, Ncv32u CN> struct __pixDemoteClampNN_CN {static __host__ __device__ Tout _pixDemoteClampNN_CN(Tin &pix);};
template<typename Tin, typename Tout> struct __pixDemoteClampNN_CN<Tin, Tout, 1> {
static __host__ __device__ Tout _pixDemoteClampNN_CN(Tin &pix)
{
Tout out;
_TDemoteClampNN(pix.x, out.x);
return out;
}};
template<typename Tin, typename Tout> struct __pixDemoteClampNN_CN<Tin, Tout, 3> {
static __host__ __device__ Tout _pixDemoteClampNN_CN(Tin &pix)
{
Tout out;
_TDemoteClampNN(pix.x, out.x);
_TDemoteClampNN(pix.y, out.y);
_TDemoteClampNN(pix.z, out.z);
return out;
}};
template<typename Tin, typename Tout> struct __pixDemoteClampNN_CN<Tin, Tout, 4> {
static __host__ __device__ Tout _pixDemoteClampNN_CN(Tin &pix)
{
Tout out;
_TDemoteClampNN(pix.x, out.x);
_TDemoteClampNN(pix.y, out.y);
_TDemoteClampNN(pix.z, out.z);
_TDemoteClampNN(pix.w, out.w);
return out;
}};
template<typename Tin, typename Tout> inline __host__ __device__ Tout _pixDemoteClampNN(Tin &pix)
{
return __pixDemoteClampNN_CN<Tin, Tout, NC(Tin)>::_pixDemoteClampNN_CN(pix);
}
template<typename Tin, typename Tout, typename Tw, Ncv32u CN> struct __pixScale_CN {static __host__ __device__ Tout _pixScale_CN(Tin &pix, Tw w);};
template<typename Tin, typename Tout, typename Tw> struct __pixScale_CN<Tin, Tout, Tw, 1> {
static __host__ __device__ Tout _pixScale_CN(Tin &pix, Tw w)
{
Tout out;
typedef typename TConvVec2Base<Tout>::TBase TBout;
out.x = (TBout)(pix.x * w);
return out;
}};
template<typename Tin, typename Tout, typename Tw> struct __pixScale_CN<Tin, Tout, Tw, 3> {
static __host__ __device__ Tout _pixScale_CN(Tin &pix, Tw w)
{
Tout out;
typedef typename TConvVec2Base<Tout>::TBase TBout;
out.x = (TBout)(pix.x * w);
out.y = (TBout)(pix.y * w);
out.z = (TBout)(pix.z * w);
return out;
}};
template<typename Tin, typename Tout, typename Tw> struct __pixScale_CN<Tin, Tout, Tw, 4> {
static __host__ __device__ Tout _pixScale_CN(Tin &pix, Tw w)
{
Tout out;
typedef typename TConvVec2Base<Tout>::TBase TBout;
out.x = (TBout)(pix.x * w);
out.y = (TBout)(pix.y * w);
out.z = (TBout)(pix.z * w);
out.w = (TBout)(pix.w * w);
return out;
}};
template<typename Tin, typename Tout, typename Tw> static __host__ __device__ Tout _pixScale(Tin &pix, Tw w)
{
return __pixScale_CN<Tin, Tout, Tw, NC(Tin)>::_pixScale_CN(pix, w);
}
template<typename Tin, typename Tout, Ncv32u CN> struct __pixAdd_CN {static __host__ __device__ Tout _pixAdd_CN(Tout &pix1, Tin &pix2);};
template<typename Tin, typename Tout> struct __pixAdd_CN<Tin, Tout, 1> {
static __host__ __device__ Tout _pixAdd_CN(Tout &pix1, Tin &pix2)
{
Tout out;
out.x = pix1.x + pix2.x;
return out;
}};
template<typename Tin, typename Tout> struct __pixAdd_CN<Tin, Tout, 3> {
static __host__ __device__ Tout _pixAdd_CN(Tout &pix1, Tin &pix2)
{
Tout out;
out.x = pix1.x + pix2.x;
out.y = pix1.y + pix2.y;
out.z = pix1.z + pix2.z;
return out;
}};
template<typename Tin, typename Tout> struct __pixAdd_CN<Tin, Tout, 4> {
static __host__ __device__ Tout _pixAdd_CN(Tout &pix1, Tin &pix2)
{
Tout out;
out.x = pix1.x + pix2.x;
out.y = pix1.y + pix2.y;
out.z = pix1.z + pix2.z;
out.w = pix1.w + pix2.w;
return out;
}};
template<typename Tin, typename Tout> static __host__ __device__ Tout _pixAdd(Tout &pix1, Tin &pix2)
{
return __pixAdd_CN<Tin, Tout, NC(Tin)>::_pixAdd_CN(pix1, pix2);
}
template<typename Tin, typename Tout, Ncv32u CN> struct __pixDist_CN {static __host__ __device__ Tout _pixDist_CN(Tin &pix1, Tin &pix2);};
template<typename Tin, typename Tout> struct __pixDist_CN<Tin, Tout, 1> {
static __host__ __device__ Tout _pixDist_CN(Tin &pix1, Tin &pix2)
{
return Tout(SQR(pix1.x - pix2.x));
}};
template<typename Tin, typename Tout> struct __pixDist_CN<Tin, Tout, 3> {
static __host__ __device__ Tout _pixDist_CN(Tin &pix1, Tin &pix2)
{
return Tout(SQR(pix1.x - pix2.x) + SQR(pix1.y - pix2.y) + SQR(pix1.z - pix2.z));
}};
template<typename Tin, typename Tout> struct __pixDist_CN<Tin, Tout, 4> {
static __host__ __device__ Tout _pixDist_CN(Tin &pix1, Tin &pix2)
{
return Tout(SQR(pix1.x - pix2.x) + SQR(pix1.y - pix2.y) + SQR(pix1.z - pix2.z) + SQR(pix1.w - pix2.w));
}};
template<typename Tin, typename Tout> static __host__ __device__ Tout _pixDist(Tin &pix1, Tin &pix2)
{
return __pixDist_CN<Tin, Tout, NC(Tin)>::_pixDist_CN(pix1, pix2);
}
template <typename T> struct TAccPixWeighted;
template<> struct TAccPixWeighted<uchar1> {typedef double1 type;};
template<> struct TAccPixWeighted<uchar3> {typedef double3 type;};
template<> struct TAccPixWeighted<uchar4> {typedef double4Compat type;};
template<> struct TAccPixWeighted<ushort1> {typedef double1 type;};
template<> struct TAccPixWeighted<ushort3> {typedef double3 type;};
template<> struct TAccPixWeighted<ushort4> {typedef double4Compat type;};
template<> struct TAccPixWeighted<float1> {typedef double1 type;};
template<> struct TAccPixWeighted<float3> {typedef double3 type;};
template<> struct TAccPixWeighted<float4> {typedef double4Compat type;};
template<typename Tfrom> struct TAccPixDist {};
template<> struct TAccPixDist<uchar1> {typedef Ncv32u type;};
template<> struct TAccPixDist<uchar3> {typedef Ncv32u type;};
template<> struct TAccPixDist<uchar4> {typedef Ncv32u type;};
template<> struct TAccPixDist<ushort1> {typedef Ncv32u type;};
template<> struct TAccPixDist<ushort3> {typedef Ncv32u type;};
template<> struct TAccPixDist<ushort4> {typedef Ncv32u type;};
template<> struct TAccPixDist<float1> {typedef Ncv32f type;};
template<> struct TAccPixDist<float3> {typedef Ncv32f type;};
template<> struct TAccPixDist<float4> {typedef Ncv32f type;};
#endif //_ncv_pixel_operations_hpp_
modules/cudalegacy/src/cuda/NCVPyramid.cu
+621
View File
@@ -0,0 +1,621 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include <stdio.h>
#include <cuda_runtime.h>
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/cudalegacy/NCV.hpp"
#include "opencv2/cudalegacy/NCVPyramid.hpp"
#include "NCVAlg.hpp"
#include "NCVPixelOperations.hpp"
template<typename T, Ncv32u CN> struct __average4_CN {static __host__ __device__ T _average4_CN(const T &p00, const T &p01, const T &p10, const T &p11);};
template<typename T> struct __average4_CN<T, 1> {
static __host__ __device__ T _average4_CN(const T &p00, const T &p01, const T &p10, const T &p11)
{
T out;
out.x = ((Ncv32s)p00.x + p01.x + p10.x + p11.x + 2) / 4;
return out;
}};
template<> struct __average4_CN<float1, 1> {
static __host__ __device__ float1 _average4_CN(const float1 &p00, const float1 &p01, const float1 &p10, const float1 &p11)
{
float1 out;
out.x = (p00.x + p01.x + p10.x + p11.x) / 4;
return out;
}};
template<> struct __average4_CN<double1, 1> {
static __host__ __device__ double1 _average4_CN(const double1 &p00, const double1 &p01, const double1 &p10, const double1 &p11)
{
double1 out;
out.x = (p00.x + p01.x + p10.x + p11.x) / 4;
return out;
}};
template<typename T> struct __average4_CN<T, 3> {
static __host__ __device__ T _average4_CN(const T &p00, const T &p01, const T &p10, const T &p11)
{
T out;
out.x = ((Ncv32s)p00.x + p01.x + p10.x + p11.x + 2) / 4;
out.y = ((Ncv32s)p00.y + p01.y + p10.y + p11.y + 2) / 4;
out.z = ((Ncv32s)p00.z + p01.z + p10.z + p11.z + 2) / 4;
return out;
}};
template<> struct __average4_CN<float3, 3> {
static __host__ __device__ float3 _average4_CN(const float3 &p00, const float3 &p01, const float3 &p10, const float3 &p11)
{
float3 out;
out.x = (p00.x + p01.x + p10.x + p11.x) / 4;
out.y = (p00.y + p01.y + p10.y + p11.y) / 4;
out.z = (p00.z + p01.z + p10.z + p11.z) / 4;
return out;
}};
template<> struct __average4_CN<double3, 3> {
static __host__ __device__ double3 _average4_CN(const double3 &p00, const double3 &p01, const double3 &p10, const double3 &p11)
{
double3 out;
out.x = (p00.x + p01.x + p10.x + p11.x) / 4;
out.y = (p00.y + p01.y + p10.y + p11.y) / 4;
out.z = (p00.z + p01.z + p10.z + p11.z) / 4;
return out;
}};
template<typename T> struct __average4_CN<T, 4> {
static __host__ __device__ T _average4_CN(const T &p00, const T &p01, const T &p10, const T &p11)
{
T out;
out.x = ((Ncv32s)p00.x + p01.x + p10.x + p11.x + 2) / 4;
out.y = ((Ncv32s)p00.y + p01.y + p10.y + p11.y + 2) / 4;
out.z = ((Ncv32s)p00.z + p01.z + p10.z + p11.z + 2) / 4;
out.w = ((Ncv32s)p00.w + p01.w + p10.w + p11.w + 2) / 4;
return out;
}};
template<> struct __average4_CN<float4, 4> {
static __host__ __device__ float4 _average4_CN(const float4 &p00, const float4 &p01, const float4 &p10, const float4 &p11)
{
float4 out;
out.x = (p00.x + p01.x + p10.x + p11.x) / 4;
out.y = (p00.y + p01.y + p10.y + p11.y) / 4;
out.z = (p00.z + p01.z + p10.z + p11.z) / 4;
out.w = (p00.w + p01.w + p10.w + p11.w) / 4;
return out;
}};
template<> struct __average4_CN<double4Compat, 4> {
static __host__ __device__ double4Compat _average4_CN(const double4Compat&p00, const double4Compat&p01, const double4Compat&p10, const double4Compat&p11)
{
double4Compat out;
out.x = (p00.x + p01.x + p10.x + p11.x) / 4;
out.y = (p00.y + p01.y + p10.y + p11.y) / 4;
out.z = (p00.z + p01.z + p10.z + p11.z) / 4;
out.w = (p00.w + p01.w + p10.w + p11.w) / 4;
return out;
}};
template<typename T> static __host__ __device__ T _average4(const T &p00, const T &p01, const T &p10, const T &p11)
{
return __average4_CN<T, NC(T)>::_average4_CN(p00, p01, p10, p11);
}
template<typename Tin, typename Tout, Ncv32u CN> struct __lerp_CN {static __host__ __device__ Tout _lerp_CN(const Tin &a, const Tin &b, Ncv32f d);};
template<typename Tin, typename Tout> struct __lerp_CN<Tin, Tout, 1> {
static __host__ __device__ Tout _lerp_CN(const Tin &a, const Tin &b, Ncv32f d)
{
typedef typename TConvVec2Base<Tout>::TBase TB;
return _pixMake(TB(b.x * d + a.x * (1 - d)));
}};
template<typename Tin, typename Tout> struct __lerp_CN<Tin, Tout, 3> {
static __host__ __device__ Tout _lerp_CN(const Tin &a, const Tin &b, Ncv32f d)
{
typedef typename TConvVec2Base<Tout>::TBase TB;
return _pixMake(TB(b.x * d + a.x * (1 - d)),
TB(b.y * d + a.y * (1 - d)),
TB(b.z * d + a.z * (1 - d)));
}};
template<typename Tin, typename Tout> struct __lerp_CN<Tin, Tout, 4> {
static __host__ __device__ Tout _lerp_CN(const Tin &a, const Tin &b, Ncv32f d)
{
typedef typename TConvVec2Base<Tout>::TBase TB;
return _pixMake(TB(b.x * d + a.x * (1 - d)),
TB(b.y * d + a.y * (1 - d)),
TB(b.z * d + a.z * (1 - d)),
TB(b.w * d + a.w * (1 - d)));
}};
template<typename Tin, typename Tout> static __host__ __device__ Tout _lerp(const Tin &a, const Tin &b, Ncv32f d)
{
return __lerp_CN<Tin, Tout, NC(Tin)>::_lerp_CN(a, b, d);
}
template<typename T>
__global__ void kernelDownsampleX2(T *d_src,
Ncv32u srcPitch,
T *d_dst,
Ncv32u dstPitch,
NcvSize32u dstRoi)
{
Ncv32u i = blockIdx.y * blockDim.y + threadIdx.y;
Ncv32u j = blockIdx.x * blockDim.x + threadIdx.x;
if (i < dstRoi.height && j < dstRoi.width)
{
T *d_src_line1 = (T *)((Ncv8u *)d_src + (2 * i + 0) * srcPitch);
T *d_src_line2 = (T *)((Ncv8u *)d_src + (2 * i + 1) * srcPitch);
T *d_dst_line = (T *)((Ncv8u *)d_dst + i * dstPitch);
T p00 = d_src_line1[2*j+0];
T p01 = d_src_line1[2*j+1];
T p10 = d_src_line2[2*j+0];
T p11 = d_src_line2[2*j+1];
d_dst_line[j] = _average4(p00, p01, p10, p11);
}
}
namespace cv { namespace cuda { namespace device
{
namespace pyramid
{
template <typename T> void kernelDownsampleX2_gpu(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
{
dim3 bDim(16, 8);
dim3 gDim(divUp(src.cols, bDim.x), divUp(src.rows, bDim.y));
kernelDownsampleX2<<<gDim, bDim, 0, stream>>>((T*)src.data, static_cast<Ncv32u>(src.step),
(T*)dst.data, static_cast<Ncv32u>(dst.step), NcvSize32u(dst.cols, dst.rows));
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void downsampleX2(PtrStepSzb src, PtrStepSzb dst, int depth, int cn, cudaStream_t stream)
{
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{kernelDownsampleX2_gpu<uchar1> , 0 /*kernelDownsampleX2_gpu<uchar2>*/ , kernelDownsampleX2_gpu<uchar3> , kernelDownsampleX2_gpu<uchar4> },
{0 /*kernelDownsampleX2_gpu<char1>*/ , 0 /*kernelDownsampleX2_gpu<char2>*/ , 0 /*kernelDownsampleX2_gpu<char3>*/ , 0 /*kernelDownsampleX2_gpu<char4>*/ },
{kernelDownsampleX2_gpu<ushort1> , 0 /*kernelDownsampleX2_gpu<ushort2>*/, kernelDownsampleX2_gpu<ushort3> , kernelDownsampleX2_gpu<ushort4> },
{0 /*kernelDownsampleX2_gpu<short1>*/ , 0 /*kernelDownsampleX2_gpu<short2>*/ , 0 /*kernelDownsampleX2_gpu<short3>*/, 0 /*kernelDownsampleX2_gpu<short4>*/},
{0 /*kernelDownsampleX2_gpu<int1>*/ , 0 /*kernelDownsampleX2_gpu<int2>*/ , 0 /*kernelDownsampleX2_gpu<int3>*/ , 0 /*kernelDownsampleX2_gpu<int4>*/ },
{kernelDownsampleX2_gpu<float1> , 0 /*kernelDownsampleX2_gpu<float2>*/ , kernelDownsampleX2_gpu<float3> , kernelDownsampleX2_gpu<float4> }
};
const func_t func = funcs[depth][cn - 1];
CV_Assert(func != 0);
func(src, dst, stream);
}
}
}}}
template<typename T>
__global__ void kernelInterpolateFrom1(T *d_srcTop,
Ncv32u srcTopPitch,
NcvSize32u szTopRoi,
T *d_dst,
Ncv32u dstPitch,
NcvSize32u dstRoi)
{
Ncv32u i = blockIdx.y * blockDim.y + threadIdx.y;
Ncv32u j = blockIdx.x * blockDim.x + threadIdx.x;
if (i < dstRoi.height && j < dstRoi.width)
{
Ncv32f ptTopX = 1.0f * (szTopRoi.width - 1) * j / (dstRoi.width - 1);
Ncv32f ptTopY = 1.0f * (szTopRoi.height - 1) * i / (dstRoi.height - 1);
Ncv32u xl = (Ncv32u)ptTopX;
Ncv32u xh = xl+1;
Ncv32f dx = ptTopX - xl;
Ncv32u yl = (Ncv32u)ptTopY;
Ncv32u yh = yl+1;
Ncv32f dy = ptTopY - yl;
T *d_src_line1 = (T *)((Ncv8u *)d_srcTop + yl * srcTopPitch);
T *d_src_line2 = (T *)((Ncv8u *)d_srcTop + yh * srcTopPitch);
T *d_dst_line = (T *)((Ncv8u *)d_dst + i * dstPitch);
T p00, p01, p10, p11;
p00 = d_src_line1[xl];
p01 = xh < szTopRoi.width ? d_src_line1[xh] : p00;
p10 = yh < szTopRoi.height ? d_src_line2[xl] : p00;
p11 = (xh < szTopRoi.width && yh < szTopRoi.height) ? d_src_line2[xh] : p00;
typedef typename TConvBase2Vec<Ncv32f, NC(T)>::TVec TVFlt;
TVFlt m_00_01 = _lerp<T, TVFlt>(p00, p01, dx);
TVFlt m_10_11 = _lerp<T, TVFlt>(p10, p11, dx);
TVFlt mixture = _lerp<TVFlt, TVFlt>(m_00_01, m_10_11, dy);
T outPix = _pixDemoteClampZ<TVFlt, T>(mixture);
d_dst_line[j] = outPix;
}
}
namespace cv { namespace cuda { namespace device
{
namespace pyramid
{
template <typename T> void kernelInterpolateFrom1_gpu(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
{
dim3 bDim(16, 8);
dim3 gDim(divUp(dst.cols, bDim.x), divUp(dst.rows, bDim.y));
kernelInterpolateFrom1<<<gDim, bDim, 0, stream>>>((T*) src.data, static_cast<Ncv32u>(src.step), NcvSize32u(src.cols, src.rows),
(T*) dst.data, static_cast<Ncv32u>(dst.step), NcvSize32u(dst.cols, dst.rows));
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void interpolateFrom1(PtrStepSzb src, PtrStepSzb dst, int depth, int cn, cudaStream_t stream)
{
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{kernelInterpolateFrom1_gpu<uchar1> , 0 /*kernelInterpolateFrom1_gpu<uchar2>*/ , kernelInterpolateFrom1_gpu<uchar3> , kernelInterpolateFrom1_gpu<uchar4> },
{0 /*kernelInterpolateFrom1_gpu<char1>*/ , 0 /*kernelInterpolateFrom1_gpu<char2>*/ , 0 /*kernelInterpolateFrom1_gpu<char3>*/ , 0 /*kernelInterpolateFrom1_gpu<char4>*/ },
{kernelInterpolateFrom1_gpu<ushort1> , 0 /*kernelInterpolateFrom1_gpu<ushort2>*/, kernelInterpolateFrom1_gpu<ushort3> , kernelInterpolateFrom1_gpu<ushort4> },
{0 /*kernelInterpolateFrom1_gpu<short1>*/, 0 /*kernelInterpolateFrom1_gpu<short2>*/ , 0 /*kernelInterpolateFrom1_gpu<short3>*/, 0 /*kernelInterpolateFrom1_gpu<short4>*/},
{0 /*kernelInterpolateFrom1_gpu<int1>*/ , 0 /*kernelInterpolateFrom1_gpu<int2>*/ , 0 /*kernelInterpolateFrom1_gpu<int3>*/ , 0 /*kernelInterpolateFrom1_gpu<int4>*/ },
{kernelInterpolateFrom1_gpu<float1> , 0 /*kernelInterpolateFrom1_gpu<float2>*/ , kernelInterpolateFrom1_gpu<float3> , kernelInterpolateFrom1_gpu<float4> }
};
const func_t func = funcs[depth][cn - 1];
CV_Assert(func != 0);
func(src, dst, stream);
}
}
}}}
#if 0 //def _WIN32
template<typename T>
static T _interpLinear(const T &a, const T &b, Ncv32f d)
{
typedef typename TConvBase2Vec<Ncv32f, NC(T)>::TVec TVFlt;
TVFlt tmp = _lerp<T, TVFlt>(a, b, d);
return _pixDemoteClampZ<TVFlt, T>(tmp);
}
template<typename T>
static T _interpBilinear(const NCVMatrix<T> &refLayer, Ncv32f x, Ncv32f y)
{
Ncv32u xl = (Ncv32u)x;
Ncv32u xh = xl+1;
Ncv32f dx = x - xl;
Ncv32u yl = (Ncv32u)y;
Ncv32u yh = yl+1;
Ncv32f dy = y - yl;
T p00, p01, p10, p11;
p00 = refLayer.at(xl, yl);
p01 = xh < refLayer.width() ? refLayer.at(xh, yl) : p00;
p10 = yh < refLayer.height() ? refLayer.at(xl, yh) : p00;
p11 = (xh < refLayer.width() && yh < refLayer.height()) ? refLayer.at(xh, yh) : p00;
typedef typename TConvBase2Vec<Ncv32f, NC(T)>::TVec TVFlt;
TVFlt m_00_01 = _lerp<T, TVFlt>(p00, p01, dx);
TVFlt m_10_11 = _lerp<T, TVFlt>(p10, p11, dx);
TVFlt mixture = _lerp<TVFlt, TVFlt>(m_00_01, m_10_11, dy);
return _pixDemoteClampZ<TVFlt, T>(mixture);
}
template <class T>
NCVImagePyramid<T>::NCVImagePyramid(const NCVMatrix<T> &img,
Ncv8u numLayers,
INCVMemAllocator &alloc,
cudaStream_t cuStream)
{
this->_isInitialized = false;
ncvAssertPrintReturn(img.memType() == alloc.memType(), "NCVImagePyramid::ctor error", );
this->layer0 = &img;
NcvSize32u szLastLayer(img.width(), img.height());
this->nLayers = 1;
NCV_SET_SKIP_COND(alloc.isCounting());
NcvBool bDeviceCode = alloc.memType() == NCVMemoryTypeDevice;
if (numLayers == 0)
{
numLayers = 255; //it will cut-off when any of the dimensions goes 1
}
#ifdef SELF_CHECK_GPU
NCVMemNativeAllocator allocCPU(NCVMemoryTypeHostPinned, 512);
#endif
for (Ncv32u i=0; i<(Ncv32u)numLayers-1; i++)
{
NcvSize32u szCurLayer(szLastLayer.width / 2, szLastLayer.height / 2);
if (szCurLayer.width == 0 || szCurLayer.height == 0)
{
break;
}
this->pyramid.push_back(new NCVMatrixAlloc<T>(alloc, szCurLayer.width, szCurLayer.height));
ncvAssertPrintReturn(((NCVMatrixAlloc<T> *)(this->pyramid[i]))->isMemAllocated(), "NCVImagePyramid::ctor error", );
this->nLayers++;
//fill in the layer
NCV_SKIP_COND_BEGIN
const NCVMatrix<T> *prevLayer = i == 0 ? this->layer0 : this->pyramid[i-1];
NCVMatrix<T> *curLayer = this->pyramid[i];
if (bDeviceCode)
{
dim3 bDim(16, 8);
dim3 gDim(divUp(szCurLayer.width, bDim.x), divUp(szCurLayer.height, bDim.y));
kernelDownsampleX2<<<gDim, bDim, 0, cuStream>>>(prevLayer->ptr(),
prevLayer->pitch(),
curLayer->ptr(),
curLayer->pitch(),
szCurLayer);
ncvAssertPrintReturn(cudaSuccess == cudaGetLastError(), "NCVImagePyramid::ctor error", );
#ifdef SELF_CHECK_GPU
NCVMatrixAlloc<T> h_prevLayer(allocCPU, prevLayer->width(), prevLayer->height());
ncvAssertPrintReturn(h_prevLayer.isMemAllocated(), "Validation failure in NCVImagePyramid::ctor", );
NCVMatrixAlloc<T> h_curLayer(allocCPU, curLayer->width(), curLayer->height());
ncvAssertPrintReturn(h_curLayer.isMemAllocated(), "Validation failure in NCVImagePyramid::ctor", );
ncvAssertPrintReturn(NCV_SUCCESS == prevLayer->copy2D(h_prevLayer, prevLayer->size(), cuStream), "Validation failure in NCVImagePyramid::ctor", );
ncvAssertPrintReturn(NCV_SUCCESS == curLayer->copy2D(h_curLayer, curLayer->size(), cuStream), "Validation failure in NCVImagePyramid::ctor", );
ncvAssertPrintReturn(cudaSuccess == cudaStreamSynchronize(cuStream), "Validation failure in NCVImagePyramid::ctor", );
for (Ncv32u i=0; i<szCurLayer.height; i++)
{
for (Ncv32u j=0; j<szCurLayer.width; j++)
{
T p00 = h_prevLayer.at(2*j+0, 2*i+0);
T p01 = h_prevLayer.at(2*j+1, 2*i+0);
T p10 = h_prevLayer.at(2*j+0, 2*i+1);
T p11 = h_prevLayer.at(2*j+1, 2*i+1);
T outGold = _average4(p00, p01, p10, p11);
T outGPU = h_curLayer.at(j, i);
ncvAssertPrintReturn(0 == memcmp(&outGold, &outGPU, sizeof(T)), "Validation failure in NCVImagePyramid::ctor with kernelDownsampleX2", );
}
}
#endif
}
else
{
for (Ncv32u i=0; i<szCurLayer.height; i++)
{
for (Ncv32u j=0; j<szCurLayer.width; j++)
{
T p00 = prevLayer->at(2*j+0, 2*i+0);
T p01 = prevLayer->at(2*j+1, 2*i+0);
T p10 = prevLayer->at(2*j+0, 2*i+1);
T p11 = prevLayer->at(2*j+1, 2*i+1);
curLayer->at(j, i) = _average4(p00, p01, p10, p11);
}
}
}
NCV_SKIP_COND_END
szLastLayer = szCurLayer;
}
this->_isInitialized = true;
}
template <class T>
NCVImagePyramid<T>::~NCVImagePyramid()
{
}
template <class T>
NcvBool NCVImagePyramid<T>::isInitialized() const
{
return this->_isInitialized;
}
template <class T>
NCVStatus NCVImagePyramid<T>::getLayer(NCVMatrix<T> &outImg,
NcvSize32u outRoi,
NcvBool bTrilinear,
cudaStream_t cuStream) const
{
ncvAssertReturn(this->isInitialized(), NCV_UNKNOWN_ERROR);
ncvAssertReturn(outImg.memType() == this->layer0->memType(), NCV_MEM_RESIDENCE_ERROR);
ncvAssertReturn(outRoi.width <= this->layer0->width() && outRoi.height <= this->layer0->height() &&
outRoi.width > 0 && outRoi.height > 0, NCV_DIMENSIONS_INVALID);
if (outRoi.width == this->layer0->width() && outRoi.height == this->layer0->height())
{
ncvAssertReturnNcvStat(this->layer0->copy2D(outImg, NcvSize32u(this->layer0->width(), this->layer0->height()), cuStream));
return NCV_SUCCESS;
}
Ncv32f lastScale = 1.0f;
Ncv32f curScale;
const NCVMatrix<T> *lastLayer = this->layer0;
const NCVMatrix<T> *curLayer = NULL;
NcvBool bUse2Refs = false;
for (Ncv32u i=0; i<this->nLayers-1; i++)
{
curScale = lastScale * 0.5f;
curLayer = this->pyramid[i];
if (outRoi.width == curLayer->width() && outRoi.height == curLayer->height())
{
ncvAssertReturnNcvStat(this->pyramid[i]->copy2D(outImg, NcvSize32u(this->pyramid[i]->width(), this->pyramid[i]->height()), cuStream));
return NCV_SUCCESS;
}
if (outRoi.width >= curLayer->width() && outRoi.height >= curLayer->height())
{
if (outRoi.width < lastLayer->width() && outRoi.height < lastLayer->height())
{
bUse2Refs = true;
}
break;
}
lastScale = curScale;
lastLayer = curLayer;
}
bUse2Refs = bUse2Refs && bTrilinear;
NCV_SET_SKIP_COND(outImg.memType() == NCVMemoryTypeNone);
NcvBool bDeviceCode = this->layer0->memType() == NCVMemoryTypeDevice;
#ifdef SELF_CHECK_GPU
NCVMemNativeAllocator allocCPU(NCVMemoryTypeHostPinned, 512);
#endif
NCV_SKIP_COND_BEGIN
if (bDeviceCode)
{
ncvAssertReturn(bUse2Refs == false, NCV_NOT_IMPLEMENTED);
dim3 bDim(16, 8);
dim3 gDim(divUp(outRoi.width, bDim.x), divUp(outRoi.height, bDim.y));
kernelInterpolateFrom1<<<gDim, bDim, 0, cuStream>>>(lastLayer->ptr(),
lastLayer->pitch(),
lastLayer->size(),
outImg.ptr(),
outImg.pitch(),
outRoi);
ncvAssertCUDAReturn(cudaGetLastError(), NCV_CUDA_ERROR);
#ifdef SELF_CHECK_GPU
ncvSafeMatAlloc(h_lastLayer, T, allocCPU, lastLayer->width(), lastLayer->height(), NCV_ALLOCATOR_BAD_ALLOC);
ncvSafeMatAlloc(h_outImg, T, allocCPU, outImg.width(), outImg.height(), NCV_ALLOCATOR_BAD_ALLOC);
ncvAssertReturnNcvStat(lastLayer->copy2D(h_lastLayer, lastLayer->size(), cuStream));
ncvAssertReturnNcvStat(outImg.copy2D(h_outImg, outRoi, cuStream));
ncvAssertCUDAReturn(cudaStreamSynchronize(cuStream), NCV_CUDA_ERROR);
for (Ncv32u i=0; i<outRoi.height; i++)
{
for (Ncv32u j=0; j<outRoi.width; j++)
{
NcvSize32u szTopLayer(lastLayer->width(), lastLayer->height());
Ncv32f ptTopX = 1.0f * (szTopLayer.width - 1) * j / (outRoi.width - 1);
Ncv32f ptTopY = 1.0f * (szTopLayer.height - 1) * i / (outRoi.height - 1);
T outGold = _interpBilinear(h_lastLayer, ptTopX, ptTopY);
ncvAssertPrintReturn(0 == memcmp(&outGold, &h_outImg.at(j,i), sizeof(T)), "Validation failure in NCVImagePyramid::ctor with kernelInterpolateFrom1", NCV_UNKNOWN_ERROR);
}
}
#endif
}
else
{
for (Ncv32u i=0; i<outRoi.height; i++)
{
for (Ncv32u j=0; j<outRoi.width; j++)
{
//top layer pixel (always exists)
NcvSize32u szTopLayer(lastLayer->width(), lastLayer->height());
Ncv32f ptTopX = 1.0f * (szTopLayer.width - 1) * j / (outRoi.width - 1);
Ncv32f ptTopY = 1.0f * (szTopLayer.height - 1) * i / (outRoi.height - 1);
T topPix = _interpBilinear(*lastLayer, ptTopX, ptTopY);
T trilinearPix = topPix;
if (bUse2Refs)
{
//bottom layer pixel (exists only if the requested scale is greater than the smallest layer scale)
NcvSize32u szBottomLayer(curLayer->width(), curLayer->height());
Ncv32f ptBottomX = 1.0f * (szBottomLayer.width - 1) * j / (outRoi.width - 1);
Ncv32f ptBottomY = 1.0f * (szBottomLayer.height - 1) * i / (outRoi.height - 1);
T bottomPix = _interpBilinear(*curLayer, ptBottomX, ptBottomY);
Ncv32f scale = (1.0f * outRoi.width / layer0->width() + 1.0f * outRoi.height / layer0->height()) / 2;
Ncv32f dl = (scale - curScale) / (lastScale - curScale);
dl = CLAMP(dl, 0.0f, 1.0f);
trilinearPix = _interpLinear(bottomPix, topPix, dl);
}
outImg.at(j, i) = trilinearPix;
}
}
}
NCV_SKIP_COND_END
return NCV_SUCCESS;
}
template class NCVImagePyramid<uchar1>;
template class NCVImagePyramid<uchar3>;
template class NCVImagePyramid<uchar4>;
template class NCVImagePyramid<ushort1>;
template class NCVImagePyramid<ushort3>;
template class NCVImagePyramid<ushort4>;
template class NCVImagePyramid<uint1>;
template class NCVImagePyramid<uint3>;
template class NCVImagePyramid<uint4>;
template class NCVImagePyramid<float1>;
template class NCVImagePyramid<float3>;
template class NCVImagePyramid<float4>;
#endif //_WIN32
modules/cudalegacy/src/cuda/NCVRuntimeTemplates.hpp
+221
View File
@@ -0,0 +1,221 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncvruntimetemplates_hpp_
#define _ncvruntimetemplates_hpp_
#if defined _MSC_VER &&_MSC_VER >= 1200
#pragma warning( disable: 4800 )
#endif
#include <stdarg.h>
#include <vector>
////////////////////////////////////////////////////////////////////////////////
// The Loki Library
// Copyright (c) 2001 by Andrei Alexandrescu
// This code accompanies the book:
// Alexandrescu, Andrei. "Modern C++ Design: Generic Programming and Design
// Patterns Applied". Copyright (c) 2001. Addison-Wesley.
// Permission to use, copy, modify, distribute and sell this software for any
// purpose is hereby granted without fee, provided that the above copyright
// notice appear in all copies and that both that copyright notice and this
// permission notice appear in supporting documentation.
// The author or Addison-Welsey Longman make no representations about the
// suitability of this software for any purpose. It is provided "as is"
// without express or implied warranty.
// http://loki-lib.sourceforge.net/index.php?n=Main.License
////////////////////////////////////////////////////////////////////////////////
namespace Loki
{
//==============================================================================
// class NullType
// Used as a placeholder for "no type here"
// Useful as an end marker in typelists
//==============================================================================
class NullType {};
//==============================================================================
// class template Typelist
// The building block of typelists of any length
// Use it through the LOKI_TYPELIST_NN macros
// Defines nested types:
// Head (first element, a non-typelist type by convention)
// Tail (second element, can be another typelist)
//==============================================================================
template <class T, class U>
struct Typelist
{
typedef T Head;
typedef U Tail;
};
//==============================================================================
// class template Int2Type
// Converts each integral constant into a unique type
// Invocation: Int2Type<v> where v is a compile-time constant integral
// Defines 'value', an enum that evaluates to v
//==============================================================================
template <int v>
struct Int2Type
{
enum { value = v };
};
namespace TL
{
//==============================================================================
// class template TypeAt
// Finds the type at a given index in a typelist
// Invocation (TList is a typelist and index is a compile-time integral
// constant):
// TypeAt<TList, index>::Result
// returns the type in position 'index' in TList
// If you pass an out-of-bounds index, the result is a compile-time error
//==============================================================================
template <class TList, unsigned int index> struct TypeAt;
template <class Head, class Tail>
struct TypeAt<Typelist<Head, Tail>, 0>
{
typedef Head Result;
};
template <class Head, class Tail, unsigned int i>
struct TypeAt<Typelist<Head, Tail>, i>
{
typedef typename TypeAt<Tail, i - 1>::Result Result;
};
}
}
////////////////////////////////////////////////////////////////////////////////
// Runtime boolean template instance dispatcher
// Cyril Crassin <cyril.crassin@icare3d.org>
// NVIDIA, 2010
////////////////////////////////////////////////////////////////////////////////
namespace NCVRuntimeTemplateBool
{
//This struct is used to transform a list of parameters into template arguments
//The idea is to build a typelist containing the arguments
//and to pass this typelist to a user defined functor
template<typename TList, int NumArguments, class Func>
struct KernelCaller
{
//Convenience function used by the user
//Takes a variable argument list, transforms it into a list
static void call(Func *functor, ...)
{
//Vector used to collect arguments
std::vector<int> templateParamList;
//Variable argument list manipulation
va_list listPointer;
va_start(listPointer, functor);
//Collect parameters into the list
for(int i=0; i<NumArguments; i++)
{
int val = va_arg(listPointer, int);
templateParamList.push_back(val);
}
va_end(listPointer);
//Call the actual typelist building function
call(*functor, templateParamList);
}
//Actual function called recursively to build a typelist based
//on a list of values
static void call( Func &functor, std::vector<int> &templateParamList)
{
//Get current parameter value in the list
NcvBool val = templateParamList[templateParamList.size() - 1];
templateParamList.pop_back();
//Select the compile time value to add into the typelist
//depending on the runtime variable and make recursive call.
//Both versions are really instantiated
if (val)
{
KernelCaller<
Loki::Typelist<typename Loki::Int2Type<1>, TList >,
NumArguments-1, Func >
::call(functor, templateParamList);
}
else
{
KernelCaller<
Loki::Typelist<typename Loki::Int2Type<0>, TList >,
NumArguments-1, Func >
::call(functor, templateParamList);
}
}
};
//Specialization for 0 value left in the list
//-> actual kernel functor call
template<class TList, class Func>
struct KernelCaller<TList, 0, Func>
{
static void call(Func &functor)
{
//Call to the functor's kernel call method
functor.call(TList()); //TList instantiated to get the method template parameter resolved
}
static void call(Func &functor, std::vector<int> &templateParams)
{
CV_UNUSED(templateParams);
functor.call(TList());
}
};
}
#endif //_ncvruntimetemplates_hpp_
modules/cudalegacy/src/cuda/NPP_staging.cu
+2284
View File
File diff suppressed because it is too large Load Diff
modules/cudalegacy/src/cuda/bm.cu
+163
View File
@@ -0,0 +1,163 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/limits.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "opencv2/core/cuda/reduce.hpp"
#include <opencv2/cudev/ptr2d/texture.hpp>
using namespace cv::cuda;
using namespace cv::cuda::device;
namespace optflowbm
{
__device__ int cmpBlocks(cv::cudev::TexturePtr<uchar> texCurr, cv::cudev::TexturePtr<uchar> texPrev, int X1, int Y1, int X2, int Y2, int2 blockSize)
{
int s = 0;
for (int y = 0; y < blockSize.y; ++y)
{
for (int x = 0; x < blockSize.x; ++x)
s += ::abs(texPrev(Y1 + y, X1 + x) -texCurr(Y2 + y, X2 + x));
}
return s;
}
__global__ void calcOptFlowBM(cv::cudev::TexturePtr<uchar> texPrev, cv::cudev::TexturePtr<uchar> texCurr, PtrStepSzf velx, PtrStepf vely, const int2 blockSize, const int2 shiftSize, const bool usePrevious,
const int maxX, const int maxY, const int acceptLevel, const int escapeLevel,
const short2* ss, const int ssCount)
{
const int j = blockIdx.x * blockDim.x + threadIdx.x;
const int i = blockIdx.y * blockDim.y + threadIdx.y;
if (i >= velx.rows || j >= velx.cols)
return;
const int X1 = j * shiftSize.x;
const int Y1 = i * shiftSize.y;
const int offX = usePrevious ? __float2int_rn(velx(i, j)) : 0;
const int offY = usePrevious ? __float2int_rn(vely(i, j)) : 0;
int X2 = X1 + offX;
int Y2 = Y1 + offY;
int dist = numeric_limits<int>::max();
if (0 <= X2 && X2 <= maxX && 0 <= Y2 && Y2 <= maxY)
dist = cmpBlocks(texPrev, texCurr, X1, Y1, X2, Y2, blockSize);
int countMin = 1;
int sumx = offX;
int sumy = offY;
if (dist > acceptLevel)
{
// do brute-force search
for (int k = 0; k < ssCount; ++k)
{
const short2 ssVal = ss[k];
const int dx = offX + ssVal.x;
const int dy = offY + ssVal.y;
X2 = X1 + dx;
Y2 = Y1 + dy;
if (0 <= X2 && X2 <= maxX && 0 <= Y2 && Y2 <= maxY)
{
const int tmpDist = cmpBlocks(texPrev, texCurr, X1, Y1, X2, Y2, blockSize);
if (tmpDist < acceptLevel)
{
sumx = dx;
sumy = dy;
countMin = 1;
break;
}
if (tmpDist < dist)
{
dist = tmpDist;
sumx = dx;
sumy = dy;
countMin = 1;
}
else if (tmpDist == dist)
{
sumx += dx;
sumy += dy;
countMin++;
}
}
}
if (dist > escapeLevel)
{
sumx = offX;
sumy = offY;
countMin = 1;
}
}
velx(i, j) = static_cast<float>(sumx) / countMin;
vely(i, j) = static_cast<float>(sumy) / countMin;
}
void calc(PtrStepSzb prev, PtrStepSzb curr, PtrStepSzf velx, PtrStepSzf vely, int2 blockSize, int2 shiftSize, bool usePrevious,
int maxX, int maxY, int acceptLevel, int escapeLevel, const short2* ss, int ssCount, cudaStream_t stream)
{
cv::cudev::Texture<uchar> texPrev(prev);
cv::cudev::Texture<uchar> texCurr(curr);
const dim3 block(32, 8);
const dim3 grid(divUp(velx.cols, block.x), divUp(vely.rows, block.y));
calcOptFlowBM<<<grid, block, 0, stream>>>(texPrev, texCurr, velx, vely, blockSize, shiftSize, usePrevious, maxX, maxY, acceptLevel, escapeLevel, ss, ssCount);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
#endif // !defined CUDA_DISABLER
modules/cudalegacy/src/cuda/bm_fast.cu
+295
View File
@@ -0,0 +1,295 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/limits.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "opencv2/core/cuda/reduce.hpp"
using namespace cv::cuda;
using namespace cv::cuda::device;
namespace optflowbm_fast
{
enum
{
CTA_SIZE = 128,
TILE_COLS = 128,
TILE_ROWS = 32,
STRIDE = CTA_SIZE
};
template <typename T> __device__ __forceinline__ int calcDist(T a, T b)
{
return ::abs(a - b);
}
template <class T> struct FastOptFlowBM
{
int search_radius;
int block_radius;
int search_window;
int block_window;
PtrStepSz<T> I0;
PtrStep<T> I1;
mutable PtrStepi buffer;
FastOptFlowBM(int search_window_, int block_window_,
PtrStepSz<T> I0_, PtrStepSz<T> I1_,
PtrStepi buffer_) :
search_radius(search_window_ / 2), block_radius(block_window_ / 2),
search_window(search_window_), block_window(block_window_),
I0(I0_), I1(I1_),
buffer(buffer_)
{
}
__device__ __forceinline__ void initSums_BruteForce(int i, int j, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
{
for (int index = threadIdx.x; index < search_window * search_window; index += STRIDE)
{
dist_sums[index] = 0;
for (int tx = 0; tx < block_window; ++tx)
col_sums(tx, index) = 0;
int y = index / search_window;
int x = index - y * search_window;
int ay = i;
int ax = j;
int by = i + y - search_radius;
int bx = j + x - search_radius;
for (int tx = -block_radius; tx <= block_radius; ++tx)
{
int col_sum = 0;
for (int ty = -block_radius; ty <= block_radius; ++ty)
{
int dist = calcDist(I0(ay + ty, ax + tx), I1(by + ty, bx + tx));
dist_sums[index] += dist;
col_sum += dist;
}
col_sums(tx + block_radius, index) = col_sum;
}
up_col_sums(j, index) = col_sums(block_window - 1, index);
}
}
__device__ __forceinline__ void shiftRight_FirstRow(int i, int j, int first, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
{
for (int index = threadIdx.x; index < search_window * search_window; index += STRIDE)
{
int y = index / search_window;
int x = index - y * search_window;
int ay = i;
int ax = j + block_radius;
int by = i + y - search_radius;
int bx = j + x - search_radius + block_radius;
int col_sum = 0;
for (int ty = -block_radius; ty <= block_radius; ++ty)
col_sum += calcDist(I0(ay + ty, ax), I1(by + ty, bx));
dist_sums[index] += col_sum - col_sums(first, index);
col_sums(first, index) = col_sum;
up_col_sums(j, index) = col_sum;
}
}
__device__ __forceinline__ void shiftRight_UpSums(int i, int j, int first, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
{
int ay = i;
int ax = j + block_radius;
T a_up = I0(ay - block_radius - 1, ax);
T a_down = I0(ay + block_radius, ax);
for(int index = threadIdx.x; index < search_window * search_window; index += STRIDE)
{
int y = index / search_window;
int x = index - y * search_window;
int by = i + y - search_radius;
int bx = j + x - search_radius + block_radius;
T b_up = I1(by - block_radius - 1, bx);
T b_down = I1(by + block_radius, bx);
int col_sum = up_col_sums(j, index) + calcDist(a_down, b_down) - calcDist(a_up, b_up);
dist_sums[index] += col_sum - col_sums(first, index);
col_sums(first, index) = col_sum;
up_col_sums(j, index) = col_sum;
}
}
__device__ __forceinline__ void convolve_window(int i, int j, const int* dist_sums, float& velx, float& vely) const
{
int bestDist = numeric_limits<int>::max();
int bestInd = -1;
for (int index = threadIdx.x; index < search_window * search_window; index += STRIDE)
{
int curDist = dist_sums[index];
if (curDist < bestDist)
{
bestDist = curDist;
bestInd = index;
}
}
__shared__ int cta_dist_buffer[CTA_SIZE];
__shared__ int cta_ind_buffer[CTA_SIZE];
reduceKeyVal<CTA_SIZE>(cta_dist_buffer, bestDist, cta_ind_buffer, bestInd, threadIdx.x, less<int>());
if (threadIdx.x == 0)
{
int y = bestInd / search_window;
int x = bestInd - y * search_window;
velx = x - search_radius;
vely = y - search_radius;
}
}
__device__ __forceinline__ void operator()(PtrStepf velx, PtrStepf vely) const
{
int tbx = blockIdx.x * TILE_COLS;
int tby = blockIdx.y * TILE_ROWS;
int tex = ::min(tbx + TILE_COLS, I0.cols);
int tey = ::min(tby + TILE_ROWS, I0.rows);
PtrStepi col_sums;
col_sums.data = buffer.ptr(I0.cols + blockIdx.x * block_window) + blockIdx.y * search_window * search_window;
col_sums.step = buffer.step;
PtrStepi up_col_sums;
up_col_sums.data = buffer.data + blockIdx.y * search_window * search_window;
up_col_sums.step = buffer.step;
extern __shared__ int dist_sums[]; //search_window * search_window
int first = 0;
for (int i = tby; i < tey; ++i)
{
for (int j = tbx; j < tex; ++j)
{
__syncthreads();
if (j == tbx)
{
initSums_BruteForce(i, j, dist_sums, col_sums, up_col_sums);
first = 0;
}
else
{
if (i == tby)
shiftRight_FirstRow(i, j, first, dist_sums, col_sums, up_col_sums);
else
shiftRight_UpSums(i, j, first, dist_sums, col_sums, up_col_sums);
first = (first + 1) % block_window;
}
__syncthreads();
convolve_window(i, j, dist_sums, velx(i, j), vely(i, j));
}
}
}
};
template<typename T> __global__ void optflowbm_fast_kernel(const FastOptFlowBM<T> fbm, PtrStepf velx, PtrStepf vely)
{
fbm(velx, vely);
}
void get_buffer_size(int src_cols, int src_rows, int search_window, int block_window, int& buffer_cols, int& buffer_rows)
{
dim3 grid(divUp(src_cols, TILE_COLS), divUp(src_rows, TILE_ROWS));
buffer_cols = search_window * search_window * grid.y;
buffer_rows = src_cols + block_window * grid.x;
}
template <typename T>
void calc(PtrStepSzb I0, PtrStepSzb I1, PtrStepSzf velx, PtrStepSzf vely, PtrStepi buffer, int search_window, int block_window, cudaStream_t stream)
{
FastOptFlowBM<T> fbm(search_window, block_window, I0, I1, buffer);
dim3 block(CTA_SIZE, 1);
dim3 grid(divUp(I0.cols, TILE_COLS), divUp(I0.rows, TILE_ROWS));
size_t smem = search_window * search_window * sizeof(int);
optflowbm_fast_kernel<<<grid, block, smem, stream>>>(fbm, velx, vely);
cudaSafeCall ( cudaGetLastError () );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void calc<uchar>(PtrStepSzb I0, PtrStepSzb I1, PtrStepSzf velx, PtrStepSzf vely, PtrStepi buffer, int search_window, int block_window, cudaStream_t stream);
}
#endif // !defined CUDA_DISABLER
modules/cudalegacy/src/cuda/calib3d.cu
+193
View File
@@ -0,0 +1,193 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/transform.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "opencv2/core/cuda/reduce.hpp"
namespace cv { namespace cuda { namespace device
{
#define SOLVE_PNP_RANSAC_MAX_NUM_ITERS 200
namespace transform_points
{
__constant__ float3 crot0;
__constant__ float3 crot1;
__constant__ float3 crot2;
__constant__ float3 ctransl;
struct TransformOp : unary_function<float3, float3>
{
__device__ __forceinline__ float3 operator()(const float3& p) const
{
return make_float3(
crot0.x * p.x + crot0.y * p.y + crot0.z * p.z + ctransl.x,
crot1.x * p.x + crot1.y * p.y + crot1.z * p.z + ctransl.y,
crot2.x * p.x + crot2.y * p.y + crot2.z * p.z + ctransl.z);
}
__host__ __device__ __forceinline__ TransformOp() {}
__host__ __device__ __forceinline__ TransformOp(const TransformOp&) {}
};
void call(const PtrStepSz<float3> src, const float* rot,
const float* transl, PtrStepSz<float3> dst,
cudaStream_t stream)
{
cudaSafeCall(cudaMemcpyToSymbol(crot0, rot, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(crot1, rot + 3, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(crot2, rot + 6, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(ctransl, transl, sizeof(float) * 3));
cv::cuda::device::transform(src, dst, TransformOp(), WithOutMask(), stream);
}
} // namespace transform_points
namespace project_points
{
__constant__ float3 crot0;
__constant__ float3 crot1;
__constant__ float3 crot2;
__constant__ float3 ctransl;
__constant__ float3 cproj0;
__constant__ float3 cproj1;
struct ProjectOp : unary_function<float3, float3>
{
__device__ __forceinline__ float2 operator()(const float3& p) const
{
// Rotate and translate in 3D
float3 t = make_float3(
crot0.x * p.x + crot0.y * p.y + crot0.z * p.z + ctransl.x,
crot1.x * p.x + crot1.y * p.y + crot1.z * p.z + ctransl.y,
crot2.x * p.x + crot2.y * p.y + crot2.z * p.z + ctransl.z);
// Project on 2D plane
return make_float2(
(cproj0.x * t.x + cproj0.y * t.y) / t.z + cproj0.z,
(cproj1.x * t.x + cproj1.y * t.y) / t.z + cproj1.z);
}
__host__ __device__ __forceinline__ ProjectOp() {}
__host__ __device__ __forceinline__ ProjectOp(const ProjectOp&) {}
};
void call(const PtrStepSz<float3> src, const float* rot,
const float* transl, const float* proj, PtrStepSz<float2> dst,
cudaStream_t stream)
{
cudaSafeCall(cudaMemcpyToSymbol(crot0, rot, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(crot1, rot + 3, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(crot2, rot + 6, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(ctransl, transl, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(cproj0, proj, sizeof(float) * 3));
cudaSafeCall(cudaMemcpyToSymbol(cproj1, proj + 3, sizeof(float) * 3));
cv::cuda::device::transform(src, dst, ProjectOp(), WithOutMask(), stream);
}
} // namespace project_points
namespace solve_pnp_ransac
{
__constant__ float3 crot_matrices[SOLVE_PNP_RANSAC_MAX_NUM_ITERS * 3];
__constant__ float3 ctransl_vectors[SOLVE_PNP_RANSAC_MAX_NUM_ITERS];
int maxNumIters()
{
return SOLVE_PNP_RANSAC_MAX_NUM_ITERS;
}
__device__ __forceinline__ float sqr(float x)
{
return x * x;
}
template <int BLOCK_SIZE>
__global__ void computeHypothesisScoresKernel(
const int num_points, const float3* object, const float2* image,
const float dist_threshold, int* g_num_inliers)
{
const float3* const &rot_mat = crot_matrices + blockIdx.x * 3;
const float3 &transl_vec = ctransl_vectors[blockIdx.x];
int num_inliers = 0;
for (int i = threadIdx.x; i < num_points; i += blockDim.x)
{
float3 p = object[i];
p = make_float3(
rot_mat[0].x * p.x + rot_mat[0].y * p.y + rot_mat[0].z * p.z + transl_vec.x,
rot_mat[1].x * p.x + rot_mat[1].y * p.y + rot_mat[1].z * p.z + transl_vec.y,
rot_mat[2].x * p.x + rot_mat[2].y * p.y + rot_mat[2].z * p.z + transl_vec.z);
p.x /= p.z;
p.y /= p.z;
float2 image_p = image[i];
if (sqr(p.x - image_p.x) + sqr(p.y - image_p.y) < dist_threshold)
++num_inliers;
}
__shared__ int s_num_inliers[BLOCK_SIZE];
reduce<BLOCK_SIZE>(s_num_inliers, num_inliers, threadIdx.x, plus<int>());
if (threadIdx.x == 0)
g_num_inliers[blockIdx.x] = num_inliers;
}
void computeHypothesisScores(
const int num_hypotheses, const int num_points, const float* rot_matrices,
const float3* transl_vectors, const float3* object, const float2* image,
const float dist_threshold, int* hypothesis_scores)
{
cudaSafeCall(cudaMemcpyToSymbol(crot_matrices, rot_matrices, num_hypotheses * 3 * sizeof(float3)));
cudaSafeCall(cudaMemcpyToSymbol(ctransl_vectors, transl_vectors, num_hypotheses * sizeof(float3)));
dim3 threads(256);
dim3 grid(num_hypotheses);
computeHypothesisScoresKernel<256><<<grid, threads>>>(
num_points, object, image, dist_threshold, hypothesis_scores);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
} // namespace solvepnp_ransac
}}} // namespace cv { namespace cuda { namespace cudev
#endif /* CUDA_DISABLER */
modules/cudalegacy/src/cuda/ccomponetns.cu
+534
View File
@@ -0,0 +1,534 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include <opencv2/core/cuda/common.hpp>
#include <opencv2/core/cuda/vec_traits.hpp>
#include <opencv2/core/cuda/vec_math.hpp>
#include <opencv2/core/cuda/emulation.hpp>
#include <iostream>
#include <stdio.h>
namespace cv { namespace cuda { namespace device
{
namespace ccl
{
enum
{
WARP_SIZE = 32,
WARP_LOG = 5,
CTA_SIZE_X = 32,
CTA_SIZE_Y = 8,
STA_SIZE_MERGE_Y = 4,
STA_SIZE_MERGE_X = 32,
TPB_X = 1,
TPB_Y = 4,
TILE_COLS = CTA_SIZE_X * TPB_X,
TILE_ROWS = CTA_SIZE_Y * TPB_Y
};
template<typename T> struct IntervalsTraits
{
typedef T elem_type;
};
template<> struct IntervalsTraits<unsigned char>
{
typedef int dist_type;
enum {ch = 1};
};
template<> struct IntervalsTraits<uchar3>
{
typedef int3 dist_type;
enum {ch = 3};
};
template<> struct IntervalsTraits<uchar4>
{
typedef int4 dist_type;
enum {ch = 4};
};
template<> struct IntervalsTraits<unsigned short>
{
typedef int dist_type;
enum {ch = 1};
};
template<> struct IntervalsTraits<ushort3>
{
typedef int3 dist_type;
enum {ch = 3};
};
template<> struct IntervalsTraits<ushort4>
{
typedef int4 dist_type;
enum {ch = 4};
};
template<> struct IntervalsTraits<float>
{
typedef float dist_type;
enum {ch = 1};
};
template<> struct IntervalsTraits<int>
{
typedef int dist_type;
enum {ch = 1};
};
typedef unsigned char component;
enum Edges { UP = 1, DOWN = 2, LEFT = 4, RIGHT = 8, EMPTY = 0xF0 };
template<typename T, int CH> struct InInterval {};
template<typename T> struct InInterval<T, 1>
{
typedef typename VecTraits<T>::elem_type E;
__host__ __device__ __forceinline__ InInterval(const float4& _lo, const float4& _hi) : lo((E)(-_lo.x)), hi((E)_hi.x) { }
T lo, hi;
template<typename I> __device__ __forceinline__ bool operator() (const I& a, const I& b) const
{
I d = a - b;
return lo <= d && d <= hi;
}
};
template<typename T> struct InInterval<T, 3>
{
typedef typename VecTraits<T>::elem_type E;
__host__ __device__ __forceinline__ InInterval(const float4& _lo, const float4& _hi)
: lo (VecTraits<T>::make((E)(-_lo.x), (E)(-_lo.y), (E)(-_lo.z))), hi (VecTraits<T>::make((E)_hi.x, (E)_hi.y, (E)_hi.z)) { }
T lo, hi;
template<typename I> __device__ __forceinline__ bool operator() (const I& a, const I& b) const
{
I d = saturate_cast<I>(a - b);
return lo.x <= d.x && d.x <= hi.x &&
lo.y <= d.y && d.y <= hi.y &&
lo.z <= d.z && d.z <= hi.z;
}
};
template<typename T> struct InInterval<T, 4>
{
typedef typename VecTraits<T>::elem_type E;
__host__ __device__ __forceinline__ InInterval(const float4& _lo, const float4& _hi)
: lo (VecTraits<T>::make((E)(-_lo.x), (E)(-_lo.y), (E)(-_lo.z), (E)(-_lo.w))), hi (VecTraits<T>::make((E)_hi.x, (E)_hi.y, (E)_hi.z, (E)_hi.w)) { }
T lo, hi;
template<typename I> __device__ __forceinline__ bool operator() (const I& a, const I& b) const
{
I d = saturate_cast<I>(a - b);
return lo.x <= d.x && d.x <= hi.x &&
lo.y <= d.y && d.y <= hi.y &&
lo.z <= d.z && d.z <= hi.z &&
lo.w <= d.w && d.w <= hi.w;
}
};
template<typename T, typename F>
__global__ void computeConnectivity(const PtrStepSz<T> image, PtrStepSzb components, F connected)
{
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
if (x >= image.cols || y >= image.rows) return;
T intensity = image(y, x);
component c = 0;
if ( x > 0 && connected(intensity, image(y, x - 1)))
c |= LEFT;
if ( y > 0 && connected(intensity, image(y - 1, x)))
c |= UP;
if ( x + 1 < image.cols && connected(intensity, image(y, x + 1)))
c |= RIGHT;
if ( y + 1 < image.rows && connected(intensity, image(y + 1, x)))
c |= DOWN;
components(y, x) = c;
}
template< typename T>
void computeEdges(const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream)
{
dim3 block(CTA_SIZE_X, CTA_SIZE_Y);
dim3 grid(divUp(image.cols, block.x), divUp(image.rows, block.y));
typedef InInterval<typename IntervalsTraits<T>::dist_type, IntervalsTraits<T>::ch> Int_t;
Int_t inInt(lo, hi);
computeConnectivity<T, Int_t><<<grid, block, 0, stream>>>(static_cast<const PtrStepSz<T> >(image), edges, inInt);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void computeEdges<uchar> (const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<uchar3> (const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<uchar4> (const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<ushort> (const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<ushort3>(const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<ushort4>(const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<int> (const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
template void computeEdges<float> (const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
__global__ void lableTiles(const PtrStepSzb edges, PtrStepSzi comps)
{
int x = threadIdx.x + blockIdx.x * TILE_COLS;
int y = threadIdx.y + blockIdx.y * TILE_ROWS;
if (x >= edges.cols || y >= edges.rows) return;
//currently x is 1
int bounds = ((y + TPB_Y) < edges.rows);
__shared__ int labelsTile[TILE_ROWS][TILE_COLS];
__shared__ int edgesTile[TILE_ROWS][TILE_COLS];
int new_labels[TPB_Y][TPB_X];
int old_labels[TPB_Y][TPB_X];
#pragma unroll
for (int i = 0; i < TPB_Y; ++i)
#pragma unroll
for (int j = 0; j < TPB_X; ++j)
{
int yloc = threadIdx.y + CTA_SIZE_Y * i;
int xloc = threadIdx.x + CTA_SIZE_X * j;
component c = edges(bounds * (y + CTA_SIZE_Y * i), x + CTA_SIZE_X * j);
if (!xloc) c &= ~LEFT;
if (!yloc) c &= ~UP;
if (xloc == TILE_COLS -1) c &= ~RIGHT;
if (yloc == TILE_ROWS -1) c &= ~DOWN;
new_labels[i][j] = yloc * TILE_COLS + xloc;
edgesTile[yloc][xloc] = c;
}
for (int k = 0; ;++k)
{
//1. backup
#pragma unroll
for (int i = 0; i < TPB_Y; ++i)
#pragma unroll
for (int j = 0; j < TPB_X; ++j)
{
int yloc = threadIdx.y + CTA_SIZE_Y * i;
int xloc = threadIdx.x + CTA_SIZE_X * j;
old_labels[i][j] = new_labels[i][j];
labelsTile[yloc][xloc] = new_labels[i][j];
}
__syncthreads();
//2. compare local arrays
#pragma unroll
for (int i = 0; i < TPB_Y; ++i)
#pragma unroll
for (int j = 0; j < TPB_X; ++j)
{
int yloc = threadIdx.y + CTA_SIZE_Y * i;
int xloc = threadIdx.x + CTA_SIZE_X * j;
component c = edgesTile[yloc][xloc];
int label = new_labels[i][j];
if (c & UP)
label = ::min(label, labelsTile[yloc - 1][xloc]);
if (c & DOWN)
label = ::min(label, labelsTile[yloc + 1][xloc]);
if (c & LEFT)
label = ::min(label, labelsTile[yloc][xloc - 1]);
if (c & RIGHT)
label = ::min(label, labelsTile[yloc][xloc + 1]);
new_labels[i][j] = label;
}
__syncthreads();
//3. determine: Is any value changed?
int changed = 0;
#pragma unroll
for (int i = 0; i < TPB_Y; ++i)
#pragma unroll
for (int j = 0; j < TPB_X; ++j)
{
if (new_labels[i][j] < old_labels[i][j])
{
changed = 1;
Emulation::smem::atomicMin(&labelsTile[0][0] + old_labels[i][j], new_labels[i][j]);
}
}
changed = Emulation::syncthreadsOr(changed);
if (!changed)
break;
//4. Compact paths
const int *labels = &labelsTile[0][0];
#pragma unroll
for (int i = 0; i < TPB_Y; ++i)
#pragma unroll
for (int j = 0; j < TPB_X; ++j)
{
int label = new_labels[i][j];
while( labels[label] < label ) label = labels[label];
new_labels[i][j] = label;
}
__syncthreads();
}
#pragma unroll
for (int i = 0; i < TPB_Y; ++i)
#pragma unroll
for (int j = 0; j < TPB_X; ++j)
{
int label = new_labels[i][j];
int yloc = label / TILE_COLS;
int xloc = label - yloc * TILE_COLS;
xloc += blockIdx.x * TILE_COLS;
yloc += blockIdx.y * TILE_ROWS;
label = yloc * edges.cols + xloc;
// do it for x too.
if (y + CTA_SIZE_Y * i < comps.rows) comps(y + CTA_SIZE_Y * i, x + CTA_SIZE_X * j) = label;
}
}
__device__ __forceinline__ int root(const PtrStepSzi& comps, int label)
{
while(1)
{
int y = label / comps.cols;
int x = label - y * comps.cols;
int parent = comps(y, x);
if (label == parent) break;
label = parent;
}
return label;
}
__device__ __forceinline__ void isConnected(PtrStepSzi& comps, int l1, int l2, bool& changed)
{
int r1 = root(comps, l1);
int r2 = root(comps, l2);
if (r1 == r2) return;
int mi = ::min(r1, r2);
int ma = ::max(r1, r2);
int y = ma / comps.cols;
int x = ma - y * comps.cols;
atomicMin(&comps.ptr(y)[x], mi);
changed = true;
}
__global__ void crossMerge(const int tilesNumY, const int tilesNumX, int tileSizeY, int tileSizeX,
const PtrStepSzb edges, PtrStepSzi comps, const int yIncomplete, int xIncomplete)
{
int tid = threadIdx.y * blockDim.x + threadIdx.x;
int stride = blockDim.y * blockDim.x;
int ybegin = blockIdx.y * (tilesNumY * tileSizeY);
int yend = ybegin + tilesNumY * tileSizeY;
if (blockIdx.y == gridDim.y - 1)
{
yend -= yIncomplete * tileSizeY;
yend -= tileSizeY;
tileSizeY = (edges.rows % tileSizeY);
yend += tileSizeY;
}
int xbegin = blockIdx.x * tilesNumX * tileSizeX;
int xend = xbegin + tilesNumX * tileSizeX;
if (blockIdx.x == gridDim.x - 1)
{
if (xIncomplete) yend = ybegin;
xend -= xIncomplete * tileSizeX;
xend -= tileSizeX;
tileSizeX = (edges.cols % tileSizeX);
xend += tileSizeX;
}
if (blockIdx.y == (gridDim.y - 1) && yIncomplete)
{
xend = xbegin;
}
int tasksV = (tilesNumX - 1) * (yend - ybegin);
int tasksH = (tilesNumY - 1) * (xend - xbegin);
int total = tasksH + tasksV;
bool changed;
do
{
changed = false;
for (int taskIdx = tid; taskIdx < total; taskIdx += stride)
{
if (taskIdx < tasksH)
{
int indexH = taskIdx;
int row = indexH / (xend - xbegin);
int col = indexH - row * (xend - xbegin);
int y = ybegin + (row + 1) * tileSizeY;
int x = xbegin + col;
component e = edges( x, y);
if (e & UP)
{
int lc = comps(y,x);
int lu = comps(y - 1, x);
isConnected(comps, lc, lu, changed);
}
}
else
{
int indexV = taskIdx - tasksH;
int col = indexV / (yend - ybegin);
int row = indexV - col * (yend - ybegin);
int x = xbegin + (col + 1) * tileSizeX;
int y = ybegin + row;
component e = edges(x, y);
if (e & LEFT)
{
int lc = comps(y, x);
int ll = comps(y, x - 1);
isConnected(comps, lc, ll, changed);
}
}
}
} while (Emulation::syncthreadsOr(changed));
}
__global__ void flatten(const PtrStepSzb edges, PtrStepSzi comps)
{
int x = threadIdx.x + blockIdx.x * blockDim.x;
int y = threadIdx.y + blockIdx.y * blockDim.y;
if( x < comps.cols && y < comps.rows)
comps(y, x) = root(comps, comps(y, x));
}
enum {CC_NO_COMPACT = 0, CC_COMPACT_LABELS = 1};
void labelComponents(const PtrStepSzb& edges, PtrStepSzi comps, int flags, cudaStream_t stream)
{
CV_UNUSED(flags);
dim3 block(CTA_SIZE_X, CTA_SIZE_Y);
dim3 grid(divUp(edges.cols, TILE_COLS), divUp(edges.rows, TILE_ROWS));
lableTiles<<<grid, block, 0, stream>>>(edges, comps);
cudaSafeCall( cudaGetLastError() );
int tileSizeX = TILE_COLS, tileSizeY = TILE_ROWS;
while (grid.x > 1 || grid.y > 1)
{
dim3 mergeGrid((int)ceilf(grid.x / 2.f), (int)ceilf(grid.y / 2.f));
dim3 mergeBlock(STA_SIZE_MERGE_X, STA_SIZE_MERGE_Y);
// debug log
// std::cout << "merging: " << grid.y << " x " << grid.x << " ---> " << mergeGrid.y << " x " << mergeGrid.x << " for tiles: " << tileSizeY << " x " << tileSizeX << std::endl;
crossMerge<<<mergeGrid, mergeBlock, 0, stream>>>(2, 2, tileSizeY, tileSizeX, edges, comps, (int)ceilf(grid.y / 2.f) - grid.y / 2, (int)ceilf(grid.x / 2.f) - grid.x / 2);
tileSizeX <<= 1;
tileSizeY <<= 1;
grid = mergeGrid;
cudaSafeCall( cudaGetLastError() );
}
grid.x = divUp(edges.cols, block.x);
grid.y = divUp(edges.rows, block.y);
flatten<<<grid, block, 0, stream>>>(edges, comps);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
} } }
#endif /* CUDA_DISABLER */
modules/cudalegacy/src/cuda/fgd.cu
+801
View File
@@ -0,0 +1,801 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_math.hpp"
#include "opencv2/core/cuda/limits.hpp"
#include "opencv2/core/cuda/utility.hpp"
#include "opencv2/core/cuda/reduce.hpp"
#include "opencv2/core/cuda/functional.hpp"
#include "fgd.hpp"
using namespace cv::cuda;
using namespace cv::cuda::device;
namespace fgd
{
////////////////////////////////////////////////////////////////////////////
// calcDiffHistogram
const unsigned int UINT_BITS = 32U;
const int LOG_WARP_SIZE = 5;
const int WARP_SIZE = 1 << LOG_WARP_SIZE;
#if (__CUDA_ARCH__ < 120)
const unsigned int TAG_MASK = (1U << (UINT_BITS - LOG_WARP_SIZE)) - 1U;
#endif
const int MERGE_THREADBLOCK_SIZE = 256;
__device__ __forceinline__ void addByte(unsigned int* s_WarpHist_, unsigned int data, unsigned int threadTag)
{
#if (__CUDA_ARCH__ < 120)
volatile unsigned int* s_WarpHist = s_WarpHist_;
unsigned int count;
do
{
count = s_WarpHist[data] & TAG_MASK;
count = threadTag | (count + 1);
s_WarpHist[data] = count;
} while (s_WarpHist[data] != count);
#else
atomicInc(s_WarpHist_ + data, (unsigned int)(-1));
#endif
}
template <typename PT, typename CT>
__global__ void calcPartialHistogram(const PtrStepSz<PT> prevFrame, const PtrStep<CT> curFrame, unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2)
{
#if (__CUDA_ARCH__ < 200)
const int HISTOGRAM_WARP_COUNT = 4;
#else
const int HISTOGRAM_WARP_COUNT = 6;
#endif
const int HISTOGRAM_THREADBLOCK_SIZE = HISTOGRAM_WARP_COUNT * WARP_SIZE;
const int HISTOGRAM_THREADBLOCK_MEMORY = HISTOGRAM_WARP_COUNT * HISTOGRAM_BIN_COUNT;
//Per-warp subhistogram storage
__shared__ unsigned int s_Hist0[HISTOGRAM_THREADBLOCK_MEMORY];
__shared__ unsigned int s_Hist1[HISTOGRAM_THREADBLOCK_MEMORY];
__shared__ unsigned int s_Hist2[HISTOGRAM_THREADBLOCK_MEMORY];
//Clear shared memory storage for current threadblock before processing
#pragma unroll
for (int i = 0; i < (HISTOGRAM_THREADBLOCK_MEMORY / HISTOGRAM_THREADBLOCK_SIZE); ++i)
{
s_Hist0[threadIdx.x + i * HISTOGRAM_THREADBLOCK_SIZE] = 0;
s_Hist1[threadIdx.x + i * HISTOGRAM_THREADBLOCK_SIZE] = 0;
s_Hist2[threadIdx.x + i * HISTOGRAM_THREADBLOCK_SIZE] = 0;
}
__syncthreads();
const unsigned int warpId = threadIdx.x >> LOG_WARP_SIZE;
unsigned int* s_WarpHist0 = s_Hist0 + warpId * HISTOGRAM_BIN_COUNT;
unsigned int* s_WarpHist1 = s_Hist1 + warpId * HISTOGRAM_BIN_COUNT;
unsigned int* s_WarpHist2 = s_Hist2 + warpId * HISTOGRAM_BIN_COUNT;
const unsigned int tag = threadIdx.x << (UINT_BITS - LOG_WARP_SIZE);
const int dataCount = prevFrame.rows * prevFrame.cols;
for (unsigned int pos = blockIdx.x * HISTOGRAM_THREADBLOCK_SIZE + threadIdx.x; pos < dataCount; pos += HISTOGRAM_THREADBLOCK_SIZE * PARTIAL_HISTOGRAM_COUNT)
{
const unsigned int y = pos / prevFrame.cols;
const unsigned int x = pos % prevFrame.cols;
PT prevVal = prevFrame(y, x);
CT curVal = curFrame(y, x);
int3 diff = make_int3(
::abs(curVal.x - prevVal.x),
::abs(curVal.y - prevVal.y),
::abs(curVal.z - prevVal.z)
);
addByte(s_WarpHist0, diff.x, tag);
addByte(s_WarpHist1, diff.y, tag);
addByte(s_WarpHist2, diff.z, tag);
}
__syncthreads();
//Merge per-warp histograms into per-block and write to global memory
for (unsigned int bin = threadIdx.x; bin < HISTOGRAM_BIN_COUNT; bin += HISTOGRAM_THREADBLOCK_SIZE)
{
unsigned int sum0 = 0;
unsigned int sum1 = 0;
unsigned int sum2 = 0;
#pragma unroll
for (int i = 0; i < HISTOGRAM_WARP_COUNT; ++i)
{
#if (__CUDA_ARCH__ < 120)
sum0 += s_Hist0[bin + i * HISTOGRAM_BIN_COUNT] & TAG_MASK;
sum1 += s_Hist1[bin + i * HISTOGRAM_BIN_COUNT] & TAG_MASK;
sum2 += s_Hist2[bin + i * HISTOGRAM_BIN_COUNT] & TAG_MASK;
#else
sum0 += s_Hist0[bin + i * HISTOGRAM_BIN_COUNT];
sum1 += s_Hist1[bin + i * HISTOGRAM_BIN_COUNT];
sum2 += s_Hist2[bin + i * HISTOGRAM_BIN_COUNT];
#endif
}
partialBuf0[blockIdx.x * HISTOGRAM_BIN_COUNT + bin] = sum0;
partialBuf1[blockIdx.x * HISTOGRAM_BIN_COUNT + bin] = sum1;
partialBuf2[blockIdx.x * HISTOGRAM_BIN_COUNT + bin] = sum2;
}
}
__global__ void mergeHistogram(const unsigned int* partialBuf0, const unsigned int* partialBuf1, const unsigned int* partialBuf2, unsigned int* hist0, unsigned int* hist1, unsigned int* hist2)
{
unsigned int sum0 = 0;
unsigned int sum1 = 0;
unsigned int sum2 = 0;
#pragma unroll
for (unsigned int i = threadIdx.x; i < PARTIAL_HISTOGRAM_COUNT; i += MERGE_THREADBLOCK_SIZE)
{
sum0 += partialBuf0[blockIdx.x + i * HISTOGRAM_BIN_COUNT];
sum1 += partialBuf1[blockIdx.x + i * HISTOGRAM_BIN_COUNT];
sum2 += partialBuf2[blockIdx.x + i * HISTOGRAM_BIN_COUNT];
}
__shared__ unsigned int data0[MERGE_THREADBLOCK_SIZE];
__shared__ unsigned int data1[MERGE_THREADBLOCK_SIZE];
__shared__ unsigned int data2[MERGE_THREADBLOCK_SIZE];
plus<unsigned int> op;
reduce<MERGE_THREADBLOCK_SIZE>(smem_tuple(data0, data1, data2), thrust::tie(sum0, sum1, sum2), threadIdx.x, thrust::make_tuple(op, op, op));
if(threadIdx.x == 0)
{
hist0[blockIdx.x] = sum0;
hist1[blockIdx.x] = sum1;
hist2[blockIdx.x] = sum2;
}
}
template <typename PT, typename CT>
void calcDiffHistogram_gpu(PtrStepSzb prevFrame, PtrStepSzb curFrame,
unsigned int* hist0, unsigned int* hist1, unsigned int* hist2,
unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2,
bool cc20, cudaStream_t stream)
{
const int HISTOGRAM_WARP_COUNT = cc20 ? 6 : 4;
const int HISTOGRAM_THREADBLOCK_SIZE = HISTOGRAM_WARP_COUNT * WARP_SIZE;
calcPartialHistogram<PT, CT><<<PARTIAL_HISTOGRAM_COUNT, HISTOGRAM_THREADBLOCK_SIZE, 0, stream>>>(
(PtrStepSz<PT>)prevFrame, (PtrStepSz<CT>)curFrame, partialBuf0, partialBuf1, partialBuf2);
cudaSafeCall( cudaGetLastError() );
mergeHistogram<<<HISTOGRAM_BIN_COUNT, MERGE_THREADBLOCK_SIZE, 0, stream>>>(partialBuf0, partialBuf1, partialBuf2, hist0, hist1, hist2);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void calcDiffHistogram_gpu<uchar3, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, unsigned int* hist0, unsigned int* hist1, unsigned int* hist2, unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2, bool cc20, cudaStream_t stream);
template void calcDiffHistogram_gpu<uchar3, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, unsigned int* hist0, unsigned int* hist1, unsigned int* hist2, unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2, bool cc20, cudaStream_t stream);
template void calcDiffHistogram_gpu<uchar4, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, unsigned int* hist0, unsigned int* hist1, unsigned int* hist2, unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2, bool cc20, cudaStream_t stream);
template void calcDiffHistogram_gpu<uchar4, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, unsigned int* hist0, unsigned int* hist1, unsigned int* hist2, unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2, bool cc20, cudaStream_t stream);
/////////////////////////////////////////////////////////////////////////
// calcDiffThreshMask
template <typename PT, typename CT>
__global__ void calcDiffThreshMask(const PtrStepSz<PT> prevFrame, const PtrStep<CT> curFrame, uchar3 bestThres, PtrStepb changeMask)
{
const int y = blockIdx.y * blockDim.y + threadIdx.y;
const int x = blockIdx.x * blockDim.x + threadIdx.x;
if (y > prevFrame.rows || x > prevFrame.cols)
return;
PT prevVal = prevFrame(y, x);
CT curVal = curFrame(y, x);
int3 diff = make_int3(
::abs(curVal.x - prevVal.x),
::abs(curVal.y - prevVal.y),
::abs(curVal.z - prevVal.z)
);
if (diff.x > bestThres.x || diff.y > bestThres.y || diff.z > bestThres.z)
changeMask(y, x) = 255;
}
template <typename PT, typename CT>
void calcDiffThreshMask_gpu(PtrStepSzb prevFrame, PtrStepSzb curFrame, uchar3 bestThres, PtrStepSzb changeMask, cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(prevFrame.cols, block.x), divUp(prevFrame.rows, block.y));
calcDiffThreshMask<PT, CT><<<grid, block, 0, stream>>>((PtrStepSz<PT>)prevFrame, (PtrStepSz<CT>)curFrame, bestThres, changeMask);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void calcDiffThreshMask_gpu<uchar3, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, uchar3 bestThres, PtrStepSzb changeMask, cudaStream_t stream);
template void calcDiffThreshMask_gpu<uchar3, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, uchar3 bestThres, PtrStepSzb changeMask, cudaStream_t stream);
template void calcDiffThreshMask_gpu<uchar4, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, uchar3 bestThres, PtrStepSzb changeMask, cudaStream_t stream);
template void calcDiffThreshMask_gpu<uchar4, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, uchar3 bestThres, PtrStepSzb changeMask, cudaStream_t stream);
/////////////////////////////////////////////////////////////////////////
// bgfgClassification
__constant__ BGPixelStat c_stat;
void setBGPixelStat(const BGPixelStat& stat)
{
cudaSafeCall( cudaMemcpyToSymbol(c_stat, &stat, sizeof(BGPixelStat)) );
}
template <typename T> struct Output;
template <> struct Output<uchar3>
{
static __device__ __forceinline__ uchar3 make(uchar v0, uchar v1, uchar v2)
{
return make_uchar3(v0, v1, v2);
}
};
template <> struct Output<uchar4>
{
static __device__ __forceinline__ uchar4 make(uchar v0, uchar v1, uchar v2)
{
return make_uchar4(v0, v1, v2, 255);
}
};
template <typename PT, typename CT, typename OT>
__global__ void bgfgClassification(const PtrStepSz<PT> prevFrame, const PtrStep<CT> curFrame,
const PtrStepb Ftd, const PtrStepb Fbd, PtrStepb foreground,
int deltaC, int deltaCC, float alpha2, int N1c, int N1cc)
{
const int i = blockIdx.y * blockDim.y + threadIdx.y;
const int j = blockIdx.x * blockDim.x + threadIdx.x;
if (i > prevFrame.rows || j > prevFrame.cols)
return;
if (Fbd(i, j) || Ftd(i, j))
{
float Pb = 0.0f;
float Pv = 0.0f;
float Pvb = 0.0f;
int val = 0;
// Is it a motion pixel?
if (Ftd(i, j))
{
if (!c_stat.is_trained_dyn_model(i, j))
val = 1;
else
{
PT prevVal = prevFrame(i, j);
CT curVal = curFrame(i, j);
// Compare with stored CCt vectors:
for (int k = 0; k < N1cc && c_stat.PV_CC(i, j, k) > alpha2; ++k)
{
OT v1 = c_stat.V1_CC<OT>(i, j, k);
OT v2 = c_stat.V2_CC<OT>(i, j, k);
if (::abs(v1.x - prevVal.x) <= deltaCC &&
::abs(v1.y - prevVal.y) <= deltaCC &&
::abs(v1.z - prevVal.z) <= deltaCC &&
::abs(v2.x - curVal.x) <= deltaCC &&
::abs(v2.y - curVal.y) <= deltaCC &&
::abs(v2.z - curVal.z) <= deltaCC)
{
Pv += c_stat.PV_CC(i, j, k);
Pvb += c_stat.PVB_CC(i, j, k);
}
}
Pb = c_stat.Pbcc(i, j);
if (2 * Pvb * Pb <= Pv)
val = 1;
}
}
else if(c_stat.is_trained_st_model(i, j))
{
CT curVal = curFrame(i, j);
// Compare with stored Ct vectors:
for (int k = 0; k < N1c && c_stat.PV_C(i, j, k) > alpha2; ++k)
{
OT v = c_stat.V_C<OT>(i, j, k);
if (::abs(v.x - curVal.x) <= deltaC &&
::abs(v.y - curVal.y) <= deltaC &&
::abs(v.z - curVal.z) <= deltaC)
{
Pv += c_stat.PV_C(i, j, k);
Pvb += c_stat.PVB_C(i, j, k);
}
}
Pb = c_stat.Pbc(i, j);
if (2 * Pvb * Pb <= Pv)
val = 1;
}
// Update foreground:
foreground(i, j) = static_cast<uchar>(val);
} // end if( change detection...
}
template <typename PT, typename CT, typename OT>
void bgfgClassification_gpu(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground,
int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(prevFrame.cols, block.x), divUp(prevFrame.rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(bgfgClassification<PT, CT, OT>, cudaFuncCachePreferL1) );
bgfgClassification<PT, CT, OT><<<grid, block, 0, stream>>>((PtrStepSz<PT>)prevFrame, (PtrStepSz<CT>)curFrame,
Ftd, Fbd, foreground,
deltaC, deltaCC, alpha2, N1c, N1cc);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void bgfgClassification_gpu<uchar3, uchar3, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar3, uchar3, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar3, uchar4, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar3, uchar4, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar4, uchar3, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar4, uchar3, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar4, uchar4, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template void bgfgClassification_gpu<uchar4, uchar4, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
////////////////////////////////////////////////////////////////////////////
// updateBackgroundModel
template <typename PT, typename CT, typename OT, class PrevFramePtr2D, class CurFramePtr2D, class FtdPtr2D, class FbdPtr2D>
__global__ void updateBackgroundModel(int cols, int rows, const PrevFramePtr2D prevFrame, const CurFramePtr2D curFrame, const FtdPtr2D Ftd, const FbdPtr2D Fbd,
PtrStepb foreground, PtrStep<OT> background,
int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T)
{
const int i = blockIdx.y * blockDim.y + threadIdx.y;
const int j = blockIdx.x * blockDim.x + threadIdx.x;
if (i > rows || j > cols)
return;
const float MIN_PV = 1e-10f;
const uchar is_trained_dyn_model = c_stat.is_trained_dyn_model(i, j);
if (Ftd(i, j) || !is_trained_dyn_model)
{
const float alpha = is_trained_dyn_model ? alpha2 : alpha3;
float Pbcc = c_stat.Pbcc(i, j);
//update Pb
Pbcc *= (1.0f - alpha);
if (!foreground(i, j))
{
Pbcc += alpha;
}
int min_dist = numeric_limits<int>::max();
int indx = -1;
PT prevVal = prevFrame(i, j);
CT curVal = curFrame(i, j);
// Find best Vi match:
for (int k = 0; k < N2cc; ++k)
{
float PV_CC = c_stat.PV_CC(i, j, k);
if (!PV_CC)
break;
if (PV_CC < MIN_PV)
{
c_stat.PV_CC(i, j, k) = 0;
c_stat.PVB_CC(i, j, k) = 0;
continue;
}
c_stat.PV_CC(i, j, k) = PV_CC * (1.0f - alpha);
c_stat.PVB_CC(i, j, k) = c_stat.PVB_CC(i, j, k) * (1.0f - alpha);
OT v1 = c_stat.V1_CC<OT>(i, j, k);
int3 val1 = make_int3(
::abs(v1.x - prevVal.x),
::abs(v1.y - prevVal.y),
::abs(v1.z - prevVal.z)
);
OT v2 = c_stat.V2_CC<OT>(i, j, k);
int3 val2 = make_int3(
::abs(v2.x - curVal.x),
::abs(v2.y - curVal.y),
::abs(v2.z - curVal.z)
);
int dist = val1.x + val1.y + val1.z + val2.x + val2.y + val2.z;
if (dist < min_dist &&
val1.x <= deltaCC && val1.y <= deltaCC && val1.z <= deltaCC &&
val2.x <= deltaCC && val2.y <= deltaCC && val2.z <= deltaCC)
{
min_dist = dist;
indx = k;
}
}
if (indx < 0)
{
// Replace N2th elem in the table by new feature:
indx = N2cc - 1;
c_stat.PV_CC(i, j, indx) = alpha;
c_stat.PVB_CC(i, j, indx) = alpha;
//udate Vt
c_stat.V1_CC<OT>(i, j, indx) = Output<OT>::make(prevVal.x, prevVal.y, prevVal.z);
c_stat.V2_CC<OT>(i, j, indx) = Output<OT>::make(curVal.x, curVal.y, curVal.z);
}
else
{
// Update:
c_stat.PV_CC(i, j, indx) += alpha;
if (!foreground(i, j))
{
c_stat.PVB_CC(i, j, indx) += alpha;
}
}
//re-sort CCt table by Pv
const float PV_CC_indx = c_stat.PV_CC(i, j, indx);
const float PVB_CC_indx = c_stat.PVB_CC(i, j, indx);
const OT V1_CC_indx = c_stat.V1_CC<OT>(i, j, indx);
const OT V2_CC_indx = c_stat.V2_CC<OT>(i, j, indx);
for (int k = 0; k < indx; ++k)
{
if (c_stat.PV_CC(i, j, k) <= PV_CC_indx)
{
//shift elements
float Pv_tmp1;
float Pv_tmp2 = PV_CC_indx;
float Pvb_tmp1;
float Pvb_tmp2 = PVB_CC_indx;
OT v1_tmp1;
OT v1_tmp2 = V1_CC_indx;
OT v2_tmp1;
OT v2_tmp2 = V2_CC_indx;
for (int l = k; l <= indx; ++l)
{
Pv_tmp1 = c_stat.PV_CC(i, j, l);
c_stat.PV_CC(i, j, l) = Pv_tmp2;
Pv_tmp2 = Pv_tmp1;
Pvb_tmp1 = c_stat.PVB_CC(i, j, l);
c_stat.PVB_CC(i, j, l) = Pvb_tmp2;
Pvb_tmp2 = Pvb_tmp1;
v1_tmp1 = c_stat.V1_CC<OT>(i, j, l);
c_stat.V1_CC<OT>(i, j, l) = v1_tmp2;
v1_tmp2 = v1_tmp1;
v2_tmp1 = c_stat.V2_CC<OT>(i, j, l);
c_stat.V2_CC<OT>(i, j, l) = v2_tmp2;
v2_tmp2 = v2_tmp1;
}
break;
}
}
float sum1 = 0.0f;
float sum2 = 0.0f;
//check "once-off" changes
for (int k = 0; k < N1cc; ++k)
{
const float PV_CC = c_stat.PV_CC(i, j, k);
if (!PV_CC)
break;
sum1 += PV_CC;
sum2 += c_stat.PVB_CC(i, j, k);
}
if (sum1 > T)
c_stat.is_trained_dyn_model(i, j) = 1;
float diff = sum1 - Pbcc * sum2;
// Update stat table:
if (diff > T)
{
//new BG features are discovered
for (int k = 0; k < N1cc; ++k)
{
const float PV_CC = c_stat.PV_CC(i, j, k);
if (!PV_CC)
break;
c_stat.PVB_CC(i, j, k) = (PV_CC - Pbcc * c_stat.PVB_CC(i, j, k)) / (1.0f - Pbcc);
}
}
c_stat.Pbcc(i, j) = Pbcc;
}
// Handle "stationary" pixel:
if (!Ftd(i, j))
{
const float alpha = c_stat.is_trained_st_model(i, j) ? alpha2 : alpha3;
float Pbc = c_stat.Pbc(i, j);
//update Pb
Pbc *= (1.0f - alpha);
if (!foreground(i, j))
{
Pbc += alpha;
}
int min_dist = numeric_limits<int>::max();
int indx = -1;
CT curVal = curFrame(i, j);
//find best Vi match
for (int k = 0; k < N2c; ++k)
{
float PV_C = c_stat.PV_C(i, j, k);
if (PV_C < MIN_PV)
{
c_stat.PV_C(i, j, k) = 0;
c_stat.PVB_C(i, j, k) = 0;
continue;
}
// Exponential decay of memory
c_stat.PV_C(i, j, k) = PV_C * (1.0f - alpha);
c_stat.PVB_C(i, j, k) = c_stat.PVB_C(i, j, k) * (1.0f - alpha);
OT v = c_stat.V_C<OT>(i, j, k);
int3 val = make_int3(
::abs(v.x - curVal.x),
::abs(v.y - curVal.y),
::abs(v.z - curVal.z)
);
int dist = val.x + val.y + val.z;
if (dist < min_dist && val.x <= deltaC && val.y <= deltaC && val.z <= deltaC)
{
min_dist = dist;
indx = k;
}
}
if (indx < 0)
{
//N2th elem in the table is replaced by a new features
indx = N2c - 1;
c_stat.PV_C(i, j, indx) = alpha;
c_stat.PVB_C(i, j, indx) = alpha;
//udate Vt
c_stat.V_C<OT>(i, j, indx) = Output<OT>::make(curVal.x, curVal.y, curVal.z);
}
else
{
//update
c_stat.PV_C(i, j, indx) += alpha;
if (!foreground(i, j))
{
c_stat.PVB_C(i, j, indx) += alpha;
}
}
//re-sort Ct table by Pv
const float PV_C_indx = c_stat.PV_C(i, j, indx);
const float PVB_C_indx = c_stat.PVB_C(i, j, indx);
OT V_C_indx = c_stat.V_C<OT>(i, j, indx);
for (int k = 0; k < indx; ++k)
{
if (c_stat.PV_C(i, j, k) <= PV_C_indx)
{
//shift elements
float Pv_tmp1;
float Pv_tmp2 = PV_C_indx;
float Pvb_tmp1;
float Pvb_tmp2 = PVB_C_indx;
OT v_tmp1;
OT v_tmp2 = V_C_indx;
for (int l = k; l <= indx; ++l)
{
Pv_tmp1 = c_stat.PV_C(i, j, l);
c_stat.PV_C(i, j, l) = Pv_tmp2;
Pv_tmp2 = Pv_tmp1;
Pvb_tmp1 = c_stat.PVB_C(i, j, l);
c_stat.PVB_C(i, j, l) = Pvb_tmp2;
Pvb_tmp2 = Pvb_tmp1;
v_tmp1 = c_stat.V_C<OT>(i, j, l);
c_stat.V_C<OT>(i, j, l) = v_tmp2;
v_tmp2 = v_tmp1;
}
break;
}
}
// Check "once-off" changes:
float sum1 = 0.0f;
float sum2 = 0.0f;
for (int k = 0; k < N1c; ++k)
{
const float PV_C = c_stat.PV_C(i, j, k);
if (!PV_C)
break;
sum1 += PV_C;
sum2 += c_stat.PVB_C(i, j, k);
}
if (sum1 > T)
c_stat.is_trained_st_model(i, j) = 1;
float diff = sum1 - Pbc * sum2;
// Update stat table:
if (diff > T)
{
//new BG features are discovered
for (int k = 0; k < N1c; ++k)
{
const float PV_C = c_stat.PV_C(i, j, k);
if (!PV_C)
break;
c_stat.PVB_C(i, j, k) = (PV_C - Pbc * c_stat.PVB_C(i, j, k)) / (1.0f - Pbc);
}
c_stat.Pbc(i, j) = 1.0f - Pbc;
}
else
{
c_stat.Pbc(i, j) = Pbc;
}
} // if !(change detection) at pixel (i,j)
// Update the reference BG image:
if (!foreground(i, j))
{
CT curVal = curFrame(i, j);
if (!Ftd(i, j) && !Fbd(i, j))
{
// Apply IIR filter:
OT oldVal = background(i, j);
int3 newVal = make_int3(
__float2int_rn(oldVal.x * (1.0f - alpha1) + curVal.x * alpha1),
__float2int_rn(oldVal.y * (1.0f - alpha1) + curVal.y * alpha1),
__float2int_rn(oldVal.z * (1.0f - alpha1) + curVal.z * alpha1)
);
background(i, j) = Output<OT>::make(
static_cast<uchar>(newVal.x),
static_cast<uchar>(newVal.y),
static_cast<uchar>(newVal.z)
);
}
else
{
background(i, j) = Output<OT>::make(curVal.x, curVal.y, curVal.z);
}
}
}
template <typename PT, typename CT, typename OT>
struct UpdateBackgroundModel
{
static void call(PtrStepSz<PT> prevFrame, PtrStepSz<CT> curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSz<OT> background,
int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T,
cudaStream_t stream)
{
dim3 block(32, 8);
dim3 grid(divUp(prevFrame.cols, block.x), divUp(prevFrame.rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(updateBackgroundModel<PT, CT, OT, PtrStep<PT>, PtrStep<CT>, PtrStepb, PtrStepb>, cudaFuncCachePreferL1) );
updateBackgroundModel<PT, CT, OT, PtrStep<PT>, PtrStep<CT>, PtrStepb, PtrStepb><<<grid, block, 0, stream>>>(
prevFrame.cols, prevFrame.rows,
prevFrame, curFrame,
Ftd, Fbd, foreground, background,
deltaC, deltaCC, alpha1, alpha2, alpha3, N1c, N1cc, N2c, N2cc, T);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template <typename PT, typename CT, typename OT>
void updateBackgroundModel_gpu(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background,
int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T,
cudaStream_t stream)
{
UpdateBackgroundModel<PT, CT, OT>::call(PtrStepSz<PT>(prevFrame), PtrStepSz<CT>(curFrame), Ftd, Fbd, foreground, PtrStepSz<OT>(background),
deltaC, deltaCC, alpha1, alpha2, alpha3, N1c, N1cc, N2c, N2cc, T, stream);
}
template void updateBackgroundModel_gpu<uchar3, uchar3, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar3, uchar3, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar3, uchar4, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar3, uchar4, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar4, uchar3, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar4, uchar3, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar4, uchar4, uchar3>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
template void updateBackgroundModel_gpu<uchar4, uchar4, uchar4>(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground, PtrStepSzb background, int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */
modules/cudalegacy/src/cuda/fgd.hpp
+189
View File
@@ -0,0 +1,189 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __FGD_BGFG_COMMON_HPP__
#define __FGD_BGFG_COMMON_HPP__
#include "opencv2/core/cuda_types.hpp"
namespace fgd
{
struct BGPixelStat
{
public:
#ifdef __CUDACC__
__device__ float& Pbc(int i, int j);
__device__ float& Pbcc(int i, int j);
__device__ unsigned char& is_trained_st_model(int i, int j);
__device__ unsigned char& is_trained_dyn_model(int i, int j);
__device__ float& PV_C(int i, int j, int k);
__device__ float& PVB_C(int i, int j, int k);
template <typename T> __device__ T& V_C(int i, int j, int k);
__device__ float& PV_CC(int i, int j, int k);
__device__ float& PVB_CC(int i, int j, int k);
template <typename T> __device__ T& V1_CC(int i, int j, int k);
template <typename T> __device__ T& V2_CC(int i, int j, int k);
#endif
int rows_;
unsigned char* Pbc_data_;
size_t Pbc_step_;
unsigned char* Pbcc_data_;
size_t Pbcc_step_;
unsigned char* is_trained_st_model_data_;
size_t is_trained_st_model_step_;
unsigned char* is_trained_dyn_model_data_;
size_t is_trained_dyn_model_step_;
unsigned char* ctable_Pv_data_;
size_t ctable_Pv_step_;
unsigned char* ctable_Pvb_data_;
size_t ctable_Pvb_step_;
unsigned char* ctable_v_data_;
size_t ctable_v_step_;
unsigned char* cctable_Pv_data_;
size_t cctable_Pv_step_;
unsigned char* cctable_Pvb_data_;
size_t cctable_Pvb_step_;
unsigned char* cctable_v1_data_;
size_t cctable_v1_step_;
unsigned char* cctable_v2_data_;
size_t cctable_v2_step_;
};
#ifdef __CUDACC__
__device__ __forceinline__ float& BGPixelStat::Pbc(int i, int j)
{
return *((float*)(Pbc_data_ + i * Pbc_step_) + j);
}
__device__ __forceinline__ float& BGPixelStat::Pbcc(int i, int j)
{
return *((float*)(Pbcc_data_ + i * Pbcc_step_) + j);
}
__device__ __forceinline__ unsigned char& BGPixelStat::is_trained_st_model(int i, int j)
{
return *((unsigned char*)(is_trained_st_model_data_ + i * is_trained_st_model_step_) + j);
}
__device__ __forceinline__ unsigned char& BGPixelStat::is_trained_dyn_model(int i, int j)
{
return *((unsigned char*)(is_trained_dyn_model_data_ + i * is_trained_dyn_model_step_) + j);
}
__device__ __forceinline__ float& BGPixelStat::PV_C(int i, int j, int k)
{
return *((float*)(ctable_Pv_data_ + ((k * rows_) + i) * ctable_Pv_step_) + j);
}
__device__ __forceinline__ float& BGPixelStat::PVB_C(int i, int j, int k)
{
return *((float*)(ctable_Pvb_data_ + ((k * rows_) + i) * ctable_Pvb_step_) + j);
}
template <typename T> __device__ __forceinline__ T& BGPixelStat::V_C(int i, int j, int k)
{
return *((T*)(ctable_v_data_ + ((k * rows_) + i) * ctable_v_step_) + j);
}
__device__ __forceinline__ float& BGPixelStat::PV_CC(int i, int j, int k)
{
return *((float*)(cctable_Pv_data_ + ((k * rows_) + i) * cctable_Pv_step_) + j);
}
__device__ __forceinline__ float& BGPixelStat::PVB_CC(int i, int j, int k)
{
return *((float*)(cctable_Pvb_data_ + ((k * rows_) + i) * cctable_Pvb_step_) + j);
}
template <typename T> __device__ __forceinline__ T& BGPixelStat::V1_CC(int i, int j, int k)
{
return *((T*)(cctable_v1_data_ + ((k * rows_) + i) * cctable_v1_step_) + j);
}
template <typename T> __device__ __forceinline__ T& BGPixelStat::V2_CC(int i, int j, int k)
{
return *((T*)(cctable_v2_data_ + ((k * rows_) + i) * cctable_v2_step_) + j);
}
#endif
const int PARTIAL_HISTOGRAM_COUNT = 240;
const int HISTOGRAM_BIN_COUNT = 256;
template <typename PT, typename CT>
void calcDiffHistogram_gpu(cv::cuda::PtrStepSzb prevFrame, cv::cuda::PtrStepSzb curFrame,
unsigned int* hist0, unsigned int* hist1, unsigned int* hist2,
unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2,
bool cc20, cudaStream_t stream);
template <typename PT, typename CT>
void calcDiffThreshMask_gpu(cv::cuda::PtrStepSzb prevFrame, cv::cuda::PtrStepSzb curFrame, uchar3 bestThres, cv::cuda::PtrStepSzb changeMask, cudaStream_t stream);
void setBGPixelStat(const BGPixelStat& stat);
template <typename PT, typename CT, typename OT>
void bgfgClassification_gpu(cv::cuda::PtrStepSzb prevFrame, cv::cuda::PtrStepSzb curFrame,
cv::cuda::PtrStepSzb Ftd, cv::cuda::PtrStepSzb Fbd, cv::cuda::PtrStepSzb foreground,
int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
template <typename PT, typename CT, typename OT>
void updateBackgroundModel_gpu(cv::cuda::PtrStepSzb prevFrame, cv::cuda::PtrStepSzb curFrame,
cv::cuda::PtrStepSzb Ftd, cv::cuda::PtrStepSzb Fbd, cv::cuda::PtrStepSzb foreground, cv::cuda::PtrStepSzb background,
int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T,
cudaStream_t stream);
}
#endif // __FGD_BGFG_COMMON_HPP__
modules/cudalegacy/src/cuda/gmg.cu
+258
View File
@@ -0,0 +1,258 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
#include "opencv2/core/cuda/vec_traits.hpp"
#include "opencv2/core/cuda/limits.hpp"
namespace cv { namespace cuda { namespace device {
namespace gmg
{
__constant__ int c_width;
__constant__ int c_height;
__constant__ float c_minVal;
__constant__ float c_maxVal;
__constant__ int c_quantizationLevels;
__constant__ float c_backgroundPrior;
__constant__ float c_decisionThreshold;
__constant__ int c_maxFeatures;
__constant__ int c_numInitializationFrames;
void loadConstants(int width, int height, float minVal, float maxVal, int quantizationLevels, float backgroundPrior,
float decisionThreshold, int maxFeatures, int numInitializationFrames)
{
cudaSafeCall( cudaMemcpyToSymbol(c_width, &width, sizeof(width)) );
cudaSafeCall( cudaMemcpyToSymbol(c_height, &height, sizeof(height)) );
cudaSafeCall( cudaMemcpyToSymbol(c_minVal, &minVal, sizeof(minVal)) );
cudaSafeCall( cudaMemcpyToSymbol(c_maxVal, &maxVal, sizeof(maxVal)) );
cudaSafeCall( cudaMemcpyToSymbol(c_quantizationLevels, &quantizationLevels, sizeof(quantizationLevels)) );
cudaSafeCall( cudaMemcpyToSymbol(c_backgroundPrior, &backgroundPrior, sizeof(backgroundPrior)) );
cudaSafeCall( cudaMemcpyToSymbol(c_decisionThreshold, &decisionThreshold, sizeof(decisionThreshold)) );
cudaSafeCall( cudaMemcpyToSymbol(c_maxFeatures, &maxFeatures, sizeof(maxFeatures)) );
cudaSafeCall( cudaMemcpyToSymbol(c_numInitializationFrames, &numInitializationFrames, sizeof(numInitializationFrames)) );
}
__device__ float findFeature(const int color, const PtrStepi& colors, const PtrStepf& weights, const int x, const int y, const int nfeatures)
{
for (int i = 0, fy = y; i < nfeatures; ++i, fy += c_height)
{
if (color == colors(fy, x))
return weights(fy, x);
}
// not in histogram, so return 0.
return 0.0f;
}
__device__ void normalizeHistogram(PtrStepf weights, const int x, const int y, const int nfeatures)
{
float total = 0.0f;
for (int i = 0, fy = y; i < nfeatures; ++i, fy += c_height)
total += weights(fy, x);
if (total != 0.0f)
{
for (int i = 0, fy = y; i < nfeatures; ++i, fy += c_height)
weights(fy, x) /= total;
}
}
__device__ bool insertFeature(const int color, const float weight, PtrStepi colors, PtrStepf weights, const int x, const int y, int& nfeatures)
{
for (int i = 0, fy = y; i < nfeatures; ++i, fy += c_height)
{
if (color == colors(fy, x))
{
// feature in histogram
weights(fy, x) += weight;
return false;
}
}
if (nfeatures == c_maxFeatures)
{
// discard oldest feature
int idx = -1;
float minVal = numeric_limits<float>::max();
for (int i = 0, fy = y; i < nfeatures; ++i, fy += c_height)
{
const float w = weights(fy, x);
if (w < minVal)
{
minVal = w;
idx = fy;
}
}
colors(idx, x) = color;
weights(idx, x) = weight;
return false;
}
colors(nfeatures * c_height + y, x) = color;
weights(nfeatures * c_height + y, x) = weight;
++nfeatures;
return true;
}
namespace detail
{
template <int cn> struct Quantization
{
template <typename T>
__device__ static int apply(const T& val)
{
int res = 0;
res |= static_cast<int>((val.x - c_minVal) * c_quantizationLevels / (c_maxVal - c_minVal));
res |= static_cast<int>((val.y - c_minVal) * c_quantizationLevels / (c_maxVal - c_minVal)) << 8;
res |= static_cast<int>((val.z - c_minVal) * c_quantizationLevels / (c_maxVal - c_minVal)) << 16;
return res;
}
};
template <> struct Quantization<1>
{
template <typename T>
__device__ static int apply(T val)
{
return static_cast<int>((val - c_minVal) * c_quantizationLevels / (c_maxVal - c_minVal));
}
};
}
template <typename T> struct Quantization : detail::Quantization<VecTraits<T>::cn> {};
template <typename SrcT>
__global__ void update(const PtrStep<SrcT> frame, PtrStepb fgmask, PtrStepi colors_, PtrStepf weights_, PtrStepi nfeatures_,
const int frameNum, const float learningRate, const bool updateBackgroundModel)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x >= c_width || y >= c_height)
return;
const SrcT pix = frame(y, x);
const int newFeatureColor = Quantization<SrcT>::apply(pix);
int nfeatures = nfeatures_(y, x);
if (frameNum >= c_numInitializationFrames)
{
// typical operation
const float weight = findFeature(newFeatureColor, colors_, weights_, x, y, nfeatures);
// see Godbehere, Matsukawa, Goldberg (2012) for reasoning behind this implementation of Bayes rule
const float posterior = (weight * c_backgroundPrior) / (weight * c_backgroundPrior + (1.0f - weight) * (1.0f - c_backgroundPrior));
const bool isForeground = ((1.0f - posterior) > c_decisionThreshold);
fgmask(y, x) = (uchar)(-isForeground);
// update histogram.
if (updateBackgroundModel)
{
for (int i = 0, fy = y; i < nfeatures; ++i, fy += c_height)
weights_(fy, x) *= 1.0f - learningRate;
bool inserted = insertFeature(newFeatureColor, learningRate, colors_, weights_, x, y, nfeatures);
if (inserted)
{
normalizeHistogram(weights_, x, y, nfeatures);
nfeatures_(y, x) = nfeatures;
}
}
}
else if (updateBackgroundModel)
{
// training-mode update
insertFeature(newFeatureColor, 1.0f, colors_, weights_, x, y, nfeatures);
if (frameNum == c_numInitializationFrames - 1)
normalizeHistogram(weights_, x, y, nfeatures);
}
}
template <typename SrcT>
void update_gpu(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures,
int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream)
{
const dim3 block(32, 8);
const dim3 grid(divUp(frame.cols, block.x), divUp(frame.rows, block.y));
cudaSafeCall( cudaFuncSetCacheConfig(update<SrcT>, cudaFuncCachePreferL1) );
update<SrcT><<<grid, block, 0, stream>>>((PtrStepSz<SrcT>) frame, fgmask, colors, weights, nfeatures, frameNum, learningRate, updateBackgroundModel);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template void update_gpu<uchar >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<uchar3 >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<uchar4 >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<ushort >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<ushort3>(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<ushort4>(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<float >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<float3 >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
template void update_gpu<float4 >(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures, int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
}
}}}
#endif /* CUDA_DISABLER */
modules/cudalegacy/src/cuda/needle_map.cu
+220
View File
@@ -0,0 +1,220 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "opencv2/core/cuda/common.hpp"
namespace cv { namespace cuda { namespace device
{
namespace optical_flow
{
#define NEEDLE_MAP_SCALE 16
#define NUM_VERTS_PER_ARROW 6
__global__ void NeedleMapAverageKernel(const PtrStepSzf u, const PtrStepf v, PtrStepf u_avg, PtrStepf v_avg)
{
__shared__ float smem[2 * NEEDLE_MAP_SCALE];
volatile float* u_col_sum = smem;
volatile float* v_col_sum = u_col_sum + NEEDLE_MAP_SCALE;
const int x = blockIdx.x * NEEDLE_MAP_SCALE + threadIdx.x;
const int y = blockIdx.y * NEEDLE_MAP_SCALE;
u_col_sum[threadIdx.x] = 0;
v_col_sum[threadIdx.x] = 0;
#pragma unroll
for(int i = 0; i < NEEDLE_MAP_SCALE; ++i)
{
u_col_sum[threadIdx.x] += u(::min(y + i, u.rows - 1), x);
v_col_sum[threadIdx.x] += v(::min(y + i, u.rows - 1), x);
}
if (threadIdx.x < 8)
{
// now add the column sums
const uint X = threadIdx.x;
if (X | (0xfe == 0xfe)) // bit 0 is 0
{
u_col_sum[threadIdx.x] += u_col_sum[threadIdx.x + 1];
v_col_sum[threadIdx.x] += v_col_sum[threadIdx.x + 1];
}
if (X | (0xfe == 0xfc)) // bits 0 & 1 == 0
{
u_col_sum[threadIdx.x] += u_col_sum[threadIdx.x + 2];
v_col_sum[threadIdx.x] += v_col_sum[threadIdx.x + 2];
}
if (X | (0xf8 == 0xf8))
{
u_col_sum[threadIdx.x] += u_col_sum[threadIdx.x + 4];
v_col_sum[threadIdx.x] += v_col_sum[threadIdx.x + 4];
}
if (X == 0)
{
u_col_sum[threadIdx.x] += u_col_sum[threadIdx.x + 8];
v_col_sum[threadIdx.x] += v_col_sum[threadIdx.x + 8];
}
}
if (threadIdx.x == 0)
{
const float coeff = 1.0f / (NEEDLE_MAP_SCALE * NEEDLE_MAP_SCALE);
u_col_sum[0] *= coeff;
v_col_sum[0] *= coeff;
u_avg(blockIdx.y, blockIdx.x) = u_col_sum[0];
v_avg(blockIdx.y, blockIdx.x) = v_col_sum[0];
}
}
void NeedleMapAverage_gpu(PtrStepSzf u, PtrStepSzf v, PtrStepSzf u_avg, PtrStepSzf v_avg)
{
const dim3 block(NEEDLE_MAP_SCALE);
const dim3 grid(u_avg.cols, u_avg.rows);
NeedleMapAverageKernel<<<grid, block>>>(u, v, u_avg, v_avg);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
__global__ void NeedleMapVertexKernel(const PtrStepSzf u_avg, const PtrStepf v_avg, float* vertex_data, float* color_data, float max_flow, float xscale, float yscale)
{
// test - just draw a triangle at each pixel
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
const float arrow_x = x * NEEDLE_MAP_SCALE + NEEDLE_MAP_SCALE / 2.0f;
const float arrow_y = y * NEEDLE_MAP_SCALE + NEEDLE_MAP_SCALE / 2.0f;
float3 v[NUM_VERTS_PER_ARROW];
if (x < u_avg.cols && y < u_avg.rows)
{
const float u_avg_val = u_avg(y, x);
const float v_avg_val = v_avg(y, x);
const float theta = ::atan2f(v_avg_val, u_avg_val);
float r = ::sqrtf(v_avg_val * v_avg_val + u_avg_val * u_avg_val);
r = fmin(14.0f * (r / max_flow), 14.0f);
v[0].z = 1.0f;
v[1].z = 0.7f;
v[2].z = 0.7f;
v[3].z = 0.7f;
v[4].z = 0.7f;
v[5].z = 1.0f;
v[0].x = arrow_x;
v[0].y = arrow_y;
v[5].x = arrow_x;
v[5].y = arrow_y;
v[2].x = arrow_x + r * ::cosf(theta);
v[2].y = arrow_y + r * ::sinf(theta);
v[3].x = v[2].x;
v[3].y = v[2].y;
r = ::fmin(r, 2.5f);
v[1].x = arrow_x + r * ::cosf(theta - CV_PI_F / 2.0f);
v[1].y = arrow_y + r * ::sinf(theta - CV_PI_F / 2.0f);
v[4].x = arrow_x + r * ::cosf(theta + CV_PI_F / 2.0f);
v[4].y = arrow_y + r * ::sinf(theta + CV_PI_F / 2.0f);
int indx = (y * u_avg.cols + x) * NUM_VERTS_PER_ARROW * 3;
color_data[indx] = (theta - CV_PI_F) / CV_PI_F * 180.0f;
vertex_data[indx++] = v[0].x * xscale;
vertex_data[indx++] = v[0].y * yscale;
vertex_data[indx++] = v[0].z;
color_data[indx] = (theta - CV_PI_F) / CV_PI_F * 180.0f;
vertex_data[indx++] = v[1].x * xscale;
vertex_data[indx++] = v[1].y * yscale;
vertex_data[indx++] = v[1].z;
color_data[indx] = (theta - CV_PI_F) / CV_PI_F * 180.0f;
vertex_data[indx++] = v[2].x * xscale;
vertex_data[indx++] = v[2].y * yscale;
vertex_data[indx++] = v[2].z;
color_data[indx] = (theta - CV_PI_F) / CV_PI_F * 180.0f;
vertex_data[indx++] = v[3].x * xscale;
vertex_data[indx++] = v[3].y * yscale;
vertex_data[indx++] = v[3].z;
color_data[indx] = (theta - CV_PI_F) / CV_PI_F * 180.0f;
vertex_data[indx++] = v[4].x * xscale;
vertex_data[indx++] = v[4].y * yscale;
vertex_data[indx++] = v[4].z;
color_data[indx] = (theta - CV_PI_F) / CV_PI_F * 180.0f;
vertex_data[indx++] = v[5].x * xscale;
vertex_data[indx++] = v[5].y * yscale;
vertex_data[indx++] = v[5].z;
}
}
void CreateOpticalFlowNeedleMap_gpu(PtrStepSzf u_avg, PtrStepSzf v_avg, float* vertex_buffer, float* color_data, float max_flow, float xscale, float yscale)
{
const dim3 block(16);
const dim3 grid(divUp(u_avg.cols, block.x), divUp(u_avg.rows, block.y));
NeedleMapVertexKernel<<<grid, block>>>(u_avg, v_avg, vertex_buffer, color_data, max_flow, xscale, yscale);
cudaSafeCall( cudaGetLastError() );
cudaSafeCall( cudaDeviceSynchronize() );
}
}
}}}
#endif /* CUDA_DISABLER */
modules/cudalegacy/src/fgd.cpp
+693
View File
@@ -0,0 +1,693 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined(HAVE_CUDA) || defined(CUDA_DISABLER) || !defined(HAVE_OPENCV_IMGPROC) || !defined(HAVE_OPENCV_CUDAARITHM) || !defined(HAVE_OPENCV_CUDAIMGPROC)
cv::cuda::FGDParams::FGDParams() { throw_no_cuda(); }
Ptr<cuda::BackgroundSubtractorFGD> cv::cuda::createBackgroundSubtractorFGD(const FGDParams&) { throw_no_cuda(); return Ptr<cuda::BackgroundSubtractorFGD>(); }
#else
#include "cuda/fgd.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/core/utils/logger.hpp"
/////////////////////////////////////////////////////////////////////////
// FGDParams
namespace
{
// Default parameters of foreground detection algorithm:
const int BGFG_FGD_LC = 128;
const int BGFG_FGD_N1C = 15;
const int BGFG_FGD_N2C = 25;
const int BGFG_FGD_LCC = 64;
const int BGFG_FGD_N1CC = 25;
const int BGFG_FGD_N2CC = 40;
// Background reference image update parameter:
const float BGFG_FGD_ALPHA_1 = 0.1f;
// stat model update parameter
// 0.002f ~ 1K frame(~45sec), 0.005 ~ 18sec (if 25fps and absolutely static BG)
const float BGFG_FGD_ALPHA_2 = 0.005f;
// start value for alpha parameter (to fast initiate statistic model)
const float BGFG_FGD_ALPHA_3 = 0.1f;
const float BGFG_FGD_DELTA = 2.0f;
const float BGFG_FGD_T = 0.9f;
const float BGFG_FGD_MINAREA= 15.0f;
}
cv::cuda::FGDParams::FGDParams()
{
Lc = BGFG_FGD_LC;
N1c = BGFG_FGD_N1C;
N2c = BGFG_FGD_N2C;
Lcc = BGFG_FGD_LCC;
N1cc = BGFG_FGD_N1CC;
N2cc = BGFG_FGD_N2CC;
delta = BGFG_FGD_DELTA;
alpha1 = BGFG_FGD_ALPHA_1;
alpha2 = BGFG_FGD_ALPHA_2;
alpha3 = BGFG_FGD_ALPHA_3;
T = BGFG_FGD_T;
minArea = BGFG_FGD_MINAREA;
is_obj_without_holes = true;
perform_morphing = 1;
}
/////////////////////////////////////////////////////////////////////////
// copyChannels
namespace
{
void copyChannels(const GpuMat& src, GpuMat& dst, int dst_cn = -1)
{
const int src_cn = src.channels();
if (dst_cn < 0)
dst_cn = src_cn;
cuda::ensureSizeIsEnough(src.size(), CV_MAKE_TYPE(src.depth(), dst_cn), dst);
if (src_cn == dst_cn)
{
src.copyTo(dst);
}
else
{
static const int cvt_codes[4][4] =
{
{-1, -1, COLOR_GRAY2BGR, COLOR_GRAY2BGRA},
{-1, -1, -1, -1},
{COLOR_BGR2GRAY, -1, -1, COLOR_BGR2BGRA},
{COLOR_BGRA2GRAY, -1, COLOR_BGRA2BGR, -1}
};
const int cvt_code = cvt_codes[src_cn - 1][dst_cn - 1];
CV_DbgAssert( cvt_code >= 0 );
cuda::cvtColor(src, dst, cvt_code, dst_cn);
}
}
}
/////////////////////////////////////////////////////////////////////////
// changeDetection
namespace
{
void calcDiffHistogram(const GpuMat& prevFrame, const GpuMat& curFrame, GpuMat& hist, GpuMat& histBuf)
{
typedef void (*func_t)(PtrStepSzb prevFrame, PtrStepSzb curFrame,
unsigned int* hist0, unsigned int* hist1, unsigned int* hist2,
unsigned int* partialBuf0, unsigned int* partialBuf1, unsigned int* partialBuf2,
bool cc20, cudaStream_t stream);
static const func_t funcs[4][4] =
{
{0,0,0,0},
{0,0,0,0},
{0,0,fgd::calcDiffHistogram_gpu<uchar3, uchar3>,fgd::calcDiffHistogram_gpu<uchar3, uchar4>},
{0,0,fgd::calcDiffHistogram_gpu<uchar4, uchar3>,fgd::calcDiffHistogram_gpu<uchar4, uchar4>}
};
hist.create(3, 256, CV_32SC1);
histBuf.create(3, fgd::PARTIAL_HISTOGRAM_COUNT * fgd::HISTOGRAM_BIN_COUNT, CV_32SC1);
funcs[prevFrame.channels() - 1][curFrame.channels() - 1](
prevFrame, curFrame,
hist.ptr<unsigned int>(0), hist.ptr<unsigned int>(1), hist.ptr<unsigned int>(2),
histBuf.ptr<unsigned int>(0), histBuf.ptr<unsigned int>(1), histBuf.ptr<unsigned int>(2),
deviceSupports(FEATURE_SET_COMPUTE_20), 0);
}
void calcRelativeVariance(unsigned int hist[3 * 256], double relativeVariance[3][fgd::HISTOGRAM_BIN_COUNT])
{
std::memset(relativeVariance, 0, 3 * fgd::HISTOGRAM_BIN_COUNT * sizeof(double));
for (int thres = fgd::HISTOGRAM_BIN_COUNT - 2; thres >= 0; --thres)
{
Vec3d sum(0.0, 0.0, 0.0);
Vec3d sqsum(0.0, 0.0, 0.0);
Vec3i count(0, 0, 0);
for (int j = thres; j < fgd::HISTOGRAM_BIN_COUNT; ++j)
{
sum[0] += static_cast<double>(j) * hist[j];
sqsum[0] += static_cast<double>(j * j) * hist[j];
count[0] += hist[j];
sum[1] += static_cast<double>(j) * hist[j + 256];
sqsum[1] += static_cast<double>(j * j) * hist[j + 256];
count[1] += hist[j + 256];
sum[2] += static_cast<double>(j) * hist[j + 512];
sqsum[2] += static_cast<double>(j * j) * hist[j + 512];
count[2] += hist[j + 512];
}
count[0] = std::max(count[0], 1);
count[1] = std::max(count[1], 1);
count[2] = std::max(count[2], 1);
Vec3d my(
sum[0] / count[0],
sum[1] / count[1],
sum[2] / count[2]
);
relativeVariance[0][thres] = std::sqrt(sqsum[0] / count[0] - my[0] * my[0]);
relativeVariance[1][thres] = std::sqrt(sqsum[1] / count[1] - my[1] * my[1]);
relativeVariance[2][thres] = std::sqrt(sqsum[2] / count[2] - my[2] * my[2]);
}
}
void calcDiffThreshMask(const GpuMat& prevFrame, const GpuMat& curFrame, Vec3d bestThres, GpuMat& changeMask)
{
typedef void (*func_t)(PtrStepSzb prevFrame, PtrStepSzb curFrame, uchar3 bestThres, PtrStepSzb changeMask, cudaStream_t stream);
static const func_t funcs[4][4] =
{
{0,0,0,0},
{0,0,0,0},
{0,0,fgd::calcDiffThreshMask_gpu<uchar3, uchar3>,fgd::calcDiffThreshMask_gpu<uchar3, uchar4>},
{0,0,fgd::calcDiffThreshMask_gpu<uchar4, uchar3>,fgd::calcDiffThreshMask_gpu<uchar4, uchar4>}
};
changeMask.setTo(Scalar::all(0));
funcs[prevFrame.channels() - 1][curFrame.channels() - 1](prevFrame, curFrame,
make_uchar3((uchar)bestThres[0], (uchar)bestThres[1], (uchar)bestThres[2]),
changeMask, 0);
}
// performs change detection for Foreground detection algorithm
void changeDetection(const GpuMat& prevFrame, const GpuMat& curFrame, GpuMat& changeMask, GpuMat& hist, GpuMat& histBuf)
{
calcDiffHistogram(prevFrame, curFrame, hist, histBuf);
unsigned int histData[3 * 256];
Mat h_hist(3, 256, CV_32SC1, histData);
hist.download(h_hist);
double relativeVariance[3][fgd::HISTOGRAM_BIN_COUNT];
calcRelativeVariance(histData, relativeVariance);
// Find maximum:
Vec3d bestThres(10.0, 10.0, 10.0);
for (int i = 0; i < fgd::HISTOGRAM_BIN_COUNT; ++i)
{
bestThres[0] = std::max(bestThres[0], relativeVariance[0][i]);
bestThres[1] = std::max(bestThres[1], relativeVariance[1][i]);
bestThres[2] = std::max(bestThres[2], relativeVariance[2][i]);
}
calcDiffThreshMask(prevFrame, curFrame, bestThres, changeMask);
}
}
/////////////////////////////////////////////////////////////////////////
// bgfgClassification
namespace
{
int bgfgClassification(const GpuMat& prevFrame, const GpuMat& curFrame,
const GpuMat& Ftd, const GpuMat& Fbd,
GpuMat& foreground,
const FGDParams& params, int out_cn)
{
typedef void (*func_t)(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd, PtrStepSzb foreground,
int deltaC, int deltaCC, float alpha2, int N1c, int N1cc, cudaStream_t stream);
static const func_t funcs[4][4][4] =
{
{
{0,0,0,0}, {0,0,0,0}, {0,0,0,0}, {0,0,0,0}
},
{
{0,0,0,0}, {0,0,0,0}, {0,0,0,0}, {0,0,0,0}
},
{
{0,0,0,0}, {0,0,0,0},
{0,0,fgd::bgfgClassification_gpu<uchar3, uchar3, uchar3>,fgd::bgfgClassification_gpu<uchar3, uchar3, uchar4>},
{0,0,fgd::bgfgClassification_gpu<uchar3, uchar4, uchar3>,fgd::bgfgClassification_gpu<uchar3, uchar4, uchar4>}
},
{
{0,0,0,0}, {0,0,0,0},
{0,0,fgd::bgfgClassification_gpu<uchar4, uchar3, uchar3>,fgd::bgfgClassification_gpu<uchar4, uchar3, uchar4>},
{0,0,fgd::bgfgClassification_gpu<uchar4, uchar4, uchar3>,fgd::bgfgClassification_gpu<uchar4, uchar4, uchar4>}
}
};
const int deltaC = cvRound(params.delta * 256 / params.Lc);
const int deltaCC = cvRound(params.delta * 256 / params.Lcc);
funcs[prevFrame.channels() - 1][curFrame.channels() - 1][out_cn - 1](prevFrame, curFrame, Ftd, Fbd, foreground,
deltaC, deltaCC, params.alpha2,
params.N1c, params.N1cc, 0);
int count = cuda::countNonZero(foreground);
cuda::multiply(foreground, Scalar::all(255), foreground);
return count;
}
}
/////////////////////////////////////////////////////////////////////////
// smoothForeground
#ifdef HAVE_OPENCV_CUDAFILTERS
namespace
{
void morphology(const GpuMat& src, GpuMat& dst, GpuMat& filterBrd, int brd, Ptr<cuda::Filter>& filter, Scalar brdVal)
{
cuda::copyMakeBorder(src, filterBrd, brd, brd, brd, brd, BORDER_CONSTANT, brdVal);
filter->apply(filterBrd(Rect(brd, brd, src.cols, src.rows)), dst);
}
void smoothForeground(GpuMat& foreground, GpuMat& filterBrd, GpuMat& buf,
Ptr<cuda::Filter>& erodeFilter, Ptr<cuda::Filter>& dilateFilter,
const FGDParams& params)
{
const int brd = params.perform_morphing;
const Scalar erodeBrdVal = Scalar::all(UCHAR_MAX);
const Scalar dilateBrdVal = Scalar::all(0);
// MORPH_OPEN
morphology(foreground, buf, filterBrd, brd, erodeFilter, erodeBrdVal);
morphology(buf, foreground, filterBrd, brd, dilateFilter, dilateBrdVal);
// MORPH_CLOSE
morphology(foreground, buf, filterBrd, brd, dilateFilter, dilateBrdVal);
morphology(buf, foreground, filterBrd, brd, erodeFilter, erodeBrdVal);
}
}
#endif
/////////////////////////////////////////////////////////////////////////
// findForegroundRegions
namespace
{
int findForegroundRegions(GpuMat& d_foreground, Mat& h_foreground, std::vector< std::vector<Point> >& foreground_regions,
const FGDParams& params)
{
int region_count = 0;
// Discard under-size foreground regions:
d_foreground.download(h_foreground);
std::vector<std::vector<cv::Point> > contours, result;
std::vector<cv::Vec4i> hierarchy;
cv::findContours(h_foreground, contours, hierarchy, cv::RETR_CCOMP, cv::CHAIN_APPROX_SIMPLE);
CV_Assert(contours.size() > 0);
// adding top-level contours to results, filtering by size
for (size_t i = 0; i < contours.size(); ++i)
{
const std::vector<cv::Point> & cnt = contours[i];
const cv::Rect brect = cv::boundingRect(cnt);
bool isHole = hierarchy[i][3] >= 0; // contour with parent is hole
if (brect.area() < params.minArea || (isHole && params.is_obj_without_holes))
continue;
foreground_regions.push_back(cnt);
}
h_foreground.setTo(0);
drawContours(h_foreground, foreground_regions, -1, Scalar::all(255), -1);
d_foreground.upload(h_foreground);
return region_count;
}
}
/////////////////////////////////////////////////////////////////////////
// updateBackgroundModel
namespace
{
void updateBackgroundModel(const GpuMat& prevFrame, const GpuMat& curFrame, const GpuMat& Ftd, const GpuMat& Fbd,
const GpuMat& foreground, GpuMat& background,
const FGDParams& params)
{
typedef void (*func_t)(PtrStepSzb prevFrame, PtrStepSzb curFrame, PtrStepSzb Ftd, PtrStepSzb Fbd,
PtrStepSzb foreground, PtrStepSzb background,
int deltaC, int deltaCC, float alpha1, float alpha2, float alpha3, int N1c, int N1cc, int N2c, int N2cc, float T, cudaStream_t stream);
static const func_t funcs[4][4][4] =
{
{
{0,0,0,0}, {0,0,0,0}, {0,0,0,0}, {0,0,0,0}
},
{
{0,0,0,0}, {0,0,0,0}, {0,0,0,0}, {0,0,0,0}
},
{
{0,0,0,0}, {0,0,0,0},
{0,0,fgd::updateBackgroundModel_gpu<uchar3, uchar3, uchar3>,fgd::updateBackgroundModel_gpu<uchar3, uchar3, uchar4>},
{0,0,fgd::updateBackgroundModel_gpu<uchar3, uchar4, uchar3>,fgd::updateBackgroundModel_gpu<uchar3, uchar4, uchar4>}
},
{
{0,0,0,0}, {0,0,0,0},
{0,0,fgd::updateBackgroundModel_gpu<uchar4, uchar3, uchar3>,fgd::updateBackgroundModel_gpu<uchar4, uchar3, uchar4>},
{0,0,fgd::updateBackgroundModel_gpu<uchar4, uchar4, uchar3>,fgd::updateBackgroundModel_gpu<uchar4, uchar4, uchar4>}
}
};
const int deltaC = cvRound(params.delta * 256 / params.Lc);
const int deltaCC = cvRound(params.delta * 256 / params.Lcc);
funcs[prevFrame.channels() - 1][curFrame.channels() - 1][background.channels() - 1](
prevFrame, curFrame, Ftd, Fbd, foreground, background,
deltaC, deltaCC, params.alpha1, params.alpha2, params.alpha3,
params.N1c, params.N1cc, params.N2c, params.N2cc, params.T,
0);
}
}
namespace
{
class BGPixelStat
{
public:
void create(Size size, const FGDParams& params);
void setTrained();
operator fgd::BGPixelStat();
private:
GpuMat Pbc_;
GpuMat Pbcc_;
GpuMat is_trained_st_model_;
GpuMat is_trained_dyn_model_;
GpuMat ctable_Pv_;
GpuMat ctable_Pvb_;
GpuMat ctable_v_;
GpuMat cctable_Pv_;
GpuMat cctable_Pvb_;
GpuMat cctable_v1_;
GpuMat cctable_v2_;
};
void BGPixelStat::create(Size size, const FGDParams& params)
{
cuda::ensureSizeIsEnough(size, CV_32FC1, Pbc_);
Pbc_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(size, CV_32FC1, Pbcc_);
Pbcc_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(size, CV_8UC1, is_trained_st_model_);
is_trained_st_model_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(size, CV_8UC1, is_trained_dyn_model_);
is_trained_dyn_model_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2c * size.height, size.width, CV_32FC1, ctable_Pv_);
ctable_Pv_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2c * size.height, size.width, CV_32FC1, ctable_Pvb_);
ctable_Pvb_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2c * size.height, size.width, CV_8UC4, ctable_v_);
ctable_v_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2cc * size.height, size.width, CV_32FC1, cctable_Pv_);
cctable_Pv_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2cc * size.height, size.width, CV_32FC1, cctable_Pvb_);
cctable_Pvb_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2cc * size.height, size.width, CV_8UC4, cctable_v1_);
cctable_v1_.setTo(Scalar::all(0));
cuda::ensureSizeIsEnough(params.N2cc * size.height, size.width, CV_8UC4, cctable_v2_);
cctable_v2_.setTo(Scalar::all(0));
}
void BGPixelStat::setTrained()
{
is_trained_st_model_.setTo(Scalar::all(1));
is_trained_dyn_model_.setTo(Scalar::all(1));
}
BGPixelStat::operator fgd::BGPixelStat()
{
fgd::BGPixelStat stat;
stat.rows_ = Pbc_.rows;
stat.Pbc_data_ = Pbc_.data;
stat.Pbc_step_ = Pbc_.step;
stat.Pbcc_data_ = Pbcc_.data;
stat.Pbcc_step_ = Pbcc_.step;
stat.is_trained_st_model_data_ = is_trained_st_model_.data;
stat.is_trained_st_model_step_ = is_trained_st_model_.step;
stat.is_trained_dyn_model_data_ = is_trained_dyn_model_.data;
stat.is_trained_dyn_model_step_ = is_trained_dyn_model_.step;
stat.ctable_Pv_data_ = ctable_Pv_.data;
stat.ctable_Pv_step_ = ctable_Pv_.step;
stat.ctable_Pvb_data_ = ctable_Pvb_.data;
stat.ctable_Pvb_step_ = ctable_Pvb_.step;
stat.ctable_v_data_ = ctable_v_.data;
stat.ctable_v_step_ = ctable_v_.step;
stat.cctable_Pv_data_ = cctable_Pv_.data;
stat.cctable_Pv_step_ = cctable_Pv_.step;
stat.cctable_Pvb_data_ = cctable_Pvb_.data;
stat.cctable_Pvb_step_ = cctable_Pvb_.step;
stat.cctable_v1_data_ = cctable_v1_.data;
stat.cctable_v1_step_ = cctable_v1_.step;
stat.cctable_v2_data_ = cctable_v2_.data;
stat.cctable_v2_step_ = cctable_v2_.step;
return stat;
}
class FGDImpl : public cuda::BackgroundSubtractorFGD
{
public:
explicit FGDImpl(const FGDParams& params);
~FGDImpl();
void apply(InputArray image, OutputArray fgmask, double learningRate=-1);
void apply(InputArray image, InputArray knownForegroundMask, OutputArray fgmask, double learningRate=-1);
void getBackgroundImage(OutputArray backgroundImage) const;
void getForegroundRegions(OutputArrayOfArrays foreground_regions);
private:
void initialize(const GpuMat& firstFrame);
FGDParams params_;
Size frameSize_;
GpuMat background_;
GpuMat foreground_;
std::vector< std::vector<Point> > foreground_regions_;
Mat h_foreground_;
GpuMat prevFrame_;
GpuMat Ftd_;
GpuMat Fbd_;
BGPixelStat stat_;
GpuMat hist_;
GpuMat histBuf_;
GpuMat buf_;
GpuMat filterBrd_;
#ifdef HAVE_OPENCV_CUDAFILTERS
Ptr<cuda::Filter> dilateFilter_;
Ptr<cuda::Filter> erodeFilter_;
#endif
};
FGDImpl::FGDImpl(const FGDParams& params) : params_(params), frameSize_(0, 0)
{
}
FGDImpl::~FGDImpl()
{
}
void FGDImpl::apply(InputArray _image, InputArray _knownForegroundMask, OutputArray _fgmask, double learningRate){
if(!_knownForegroundMask.empty())
{
CV_LOG_WARNING(NULL, "Known Foreground Masking has not been implemented for this specific background subtractor, falling back to subtraction without known foreground");
}
apply(_image, _fgmask, learningRate);
}
void FGDImpl::apply(InputArray _frame, OutputArray fgmask, double)
{
GpuMat curFrame = _frame.getGpuMat();
if (curFrame.size() != frameSize_)
{
initialize(curFrame);
return;
}
CV_Assert( curFrame.type() == CV_8UC3 || curFrame.type() == CV_8UC4 );
CV_Assert( curFrame.size() == prevFrame_.size() );
foreground_regions_.clear();
foreground_.setTo(Scalar::all(0));
changeDetection(prevFrame_, curFrame, Ftd_, hist_, histBuf_);
changeDetection(background_, curFrame, Fbd_, hist_, histBuf_);
int FG_pixels_count = bgfgClassification(prevFrame_, curFrame, Ftd_, Fbd_, foreground_, params_, 4);
#ifdef HAVE_OPENCV_CUDAFILTERS
if (params_.perform_morphing > 0)
smoothForeground(foreground_, filterBrd_, buf_, erodeFilter_, dilateFilter_, params_);
#endif
if (params_.minArea > 0 || params_.is_obj_without_holes)
findForegroundRegions(foreground_, h_foreground_, foreground_regions_, params_);
// Check ALL BG update condition:
const double BGFG_FGD_BG_UPDATE_TRESH = 0.5;
if (static_cast<double>(FG_pixels_count) / Ftd_.size().area() > BGFG_FGD_BG_UPDATE_TRESH)
stat_.setTrained();
updateBackgroundModel(prevFrame_, curFrame, Ftd_, Fbd_, foreground_, background_, params_);
copyChannels(curFrame, prevFrame_, 4);
foreground_.copyTo(fgmask);
}
void FGDImpl::getBackgroundImage(OutputArray backgroundImage) const
{
cuda::cvtColor(background_, backgroundImage, COLOR_BGRA2BGR);
}
void FGDImpl::getForegroundRegions(OutputArrayOfArrays dst)
{
size_t total = foreground_regions_.size();
dst.create((int) total, 1, 0, -1, true);
for (size_t i = 0; i < total; ++i)
{
std::vector<Point>& c = foreground_regions_[i];
dst.create((int) c.size(), 1, CV_32SC2, (int) i, true);
Mat ci = dst.getMat((int) i);
Mat(ci.size(), ci.type(), &c[0]).copyTo(ci);
}
}
void FGDImpl::initialize(const GpuMat& firstFrame)
{
CV_Assert( firstFrame.type() == CV_8UC3 || firstFrame.type() == CV_8UC4 );
frameSize_ = firstFrame.size();
cuda::ensureSizeIsEnough(firstFrame.size(), CV_8UC1, foreground_);
copyChannels(firstFrame, background_, 4);
copyChannels(firstFrame, prevFrame_, 4);
cuda::ensureSizeIsEnough(firstFrame.size(), CV_8UC1, Ftd_);
cuda::ensureSizeIsEnough(firstFrame.size(), CV_8UC1, Fbd_);
stat_.create(firstFrame.size(), params_);
fgd::setBGPixelStat(stat_);
#ifdef HAVE_OPENCV_CUDAFILTERS
if (params_.perform_morphing > 0)
{
Mat kernel = getStructuringElement(MORPH_RECT, Size(1 + params_.perform_morphing * 2, 1 + params_.perform_morphing * 2));
Point anchor(params_.perform_morphing, params_.perform_morphing);
dilateFilter_ = cuda::createMorphologyFilter(MORPH_DILATE, CV_8UC1, kernel, anchor);
erodeFilter_ = cuda::createMorphologyFilter(MORPH_ERODE, CV_8UC1, kernel, anchor);
}
#endif
}
}
Ptr<cuda::BackgroundSubtractorFGD> cv::cuda::createBackgroundSubtractorFGD(const FGDParams& params)
{
return makePtr<FGDImpl>(params);
}
#endif // HAVE_CUDA
modules/cudalegacy/src/gmg.cpp
+298
View File
@@ -0,0 +1,298 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include "opencv2/core/utils/logger.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined HAVE_CUDA || defined(CUDA_DISABLER)
Ptr<cuda::BackgroundSubtractorGMG> cv::cuda::createBackgroundSubtractorGMG(int, double) { throw_no_cuda(); return Ptr<cuda::BackgroundSubtractorGMG>(); }
#else
namespace cv { namespace cuda { namespace device {
namespace gmg
{
void loadConstants(int width, int height, float minVal, float maxVal, int quantizationLevels, float backgroundPrior,
float decisionThreshold, int maxFeatures, int numInitializationFrames);
template <typename SrcT>
void update_gpu(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures,
int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
}
}}}
namespace
{
class GMGImpl : public cuda::BackgroundSubtractorGMG
{
public:
GMGImpl(int initializationFrames, double decisionThreshold);
void apply(InputArray image, OutputArray fgmask, double learningRate=-1);
void apply(InputArray image, OutputArray fgmask, double learningRate, Stream& stream);
// Overloaded Background Subtractor Applys featuring knownForegroundMask parameter
void apply(InputArray image, InputArray knownForegroundMask, OutputArray fgmask, double learningRate=-1);
void apply(InputArray image, InputArray knownForegroundMask, OutputArray fgmask, double learningRate, Stream& stream);
void getBackgroundImage(OutputArray backgroundImage) const;
int getMaxFeatures() const { return maxFeatures_; }
void setMaxFeatures(int maxFeatures) { maxFeatures_ = maxFeatures; }
double getDefaultLearningRate() const { return learningRate_; }
void setDefaultLearningRate(double lr) { learningRate_ = (float) lr; }
int getNumFrames() const { return numInitializationFrames_; }
void setNumFrames(int nframes) { numInitializationFrames_ = nframes; }
int getQuantizationLevels() const { return quantizationLevels_; }
void setQuantizationLevels(int nlevels) { quantizationLevels_ = nlevels; }
double getBackgroundPrior() const { return backgroundPrior_; }
void setBackgroundPrior(double bgprior) { backgroundPrior_ = (float) bgprior; }
int getSmoothingRadius() const { return smoothingRadius_; }
void setSmoothingRadius(int radius) { smoothingRadius_ = radius; }
double getDecisionThreshold() const { return decisionThreshold_; }
void setDecisionThreshold(double thresh) { decisionThreshold_ = (float) thresh; }
bool getUpdateBackgroundModel() const { return updateBackgroundModel_; }
void setUpdateBackgroundModel(bool update) { updateBackgroundModel_ = update; }
double getMinVal() const { return minVal_; }
void setMinVal(double val) { minVal_ = (float) val; }
double getMaxVal() const { return maxVal_; }
void setMaxVal(double val) { maxVal_ = (float) val; }
private:
void initialize(Size frameSize, float min, float max);
//! Total number of distinct colors to maintain in histogram.
int maxFeatures_;
//! Set between 0.0 and 1.0, determines how quickly features are "forgotten" from histograms.
float learningRate_;
//! Number of frames of video to use to initialize histograms.
int numInitializationFrames_;
//! Number of discrete levels in each channel to be used in histograms.
int quantizationLevels_;
//! Prior probability that any given pixel is a background pixel. A sensitivity parameter.
float backgroundPrior_;
//! Smoothing radius, in pixels, for cleaning up FG image.
int smoothingRadius_;
//! Value above which pixel is determined to be FG.
float decisionThreshold_;
//! Perform background model update.
bool updateBackgroundModel_;
float minVal_, maxVal_;
Size frameSize_;
int frameNum_;
GpuMat nfeatures_;
GpuMat colors_;
GpuMat weights_;
#if defined(HAVE_OPENCV_CUDAFILTERS) && defined(HAVE_OPENCV_CUDAARITHM)
Ptr<cuda::Filter> boxFilter_;
GpuMat buf_;
#endif
};
GMGImpl::GMGImpl(int initializationFrames, double decisionThreshold)
{
maxFeatures_ = 64;
learningRate_ = 0.025f;
numInitializationFrames_ = initializationFrames;
quantizationLevels_ = 16;
backgroundPrior_ = 0.8f;
decisionThreshold_ = (float) decisionThreshold;
smoothingRadius_ = 7;
updateBackgroundModel_ = true;
minVal_ = maxVal_ = 0;
}
void GMGImpl::apply(InputArray image, OutputArray fgmask, double learningRate)
{
apply(image, fgmask, learningRate, Stream::Null());
}
void GMGImpl::apply(InputArray _image, InputArray _knownForegroundMask, OutputArray _fgmask, double learningRate){
if(!_knownForegroundMask.empty())
{
CV_LOG_WARNING(NULL, "Known Foreground Masking has not been implemented for this specific background subtractor, falling back to subtraction without known foreground");
}
apply(_image, _fgmask, learningRate);
}
void GMGImpl::apply(InputArray _image, InputArray _knownForegroundMask, OutputArray _fgmask, double learningRate, Stream& stream){
if(!_knownForegroundMask.empty())
{
CV_LOG_WARNING(NULL, "Known Foreground Masking has not been implemented for this specific background subtractor, falling back to subtraction without known foreground");
}
apply(_image, _fgmask, learningRate, stream);
}
void GMGImpl::apply(InputArray _frame, OutputArray _fgmask, double newLearningRate, Stream& stream)
{
using namespace cv::cuda::device::gmg;
typedef void (*func_t)(PtrStepSzb frame, PtrStepb fgmask, PtrStepSzi colors, PtrStepf weights, PtrStepi nfeatures,
int frameNum, float learningRate, bool updateBackgroundModel, cudaStream_t stream);
static const func_t funcs[6][4] =
{
{update_gpu<uchar>, 0, update_gpu<uchar3>, update_gpu<uchar4>},
{0,0,0,0},
{update_gpu<ushort>, 0, update_gpu<ushort3>, update_gpu<ushort4>},
{0,0,0,0},
{0,0,0,0},
{update_gpu<float>, 0, update_gpu<float3>, update_gpu<float4>}
};
GpuMat frame = _frame.getGpuMat();
CV_Assert( frame.depth() == CV_8U || frame.depth() == CV_16U || frame.depth() == CV_32F );
CV_Assert( frame.channels() == 1 || frame.channels() == 3 || frame.channels() == 4 );
if (newLearningRate != -1.0)
{
CV_Assert( newLearningRate >= 0.0 && newLearningRate <= 1.0 );
learningRate_ = (float) newLearningRate;
}
if (frame.size() != frameSize_)
{
double minVal = minVal_;
double maxVal = maxVal_;
if (minVal_ == 0 && maxVal_ == 0)
{
minVal = 0;
maxVal = frame.depth() == CV_8U ? 255.0 : frame.depth() == CV_16U ? std::numeric_limits<ushort>::max() : 1.0;
}
initialize(frame.size(), (float) minVal, (float) maxVal);
}
_fgmask.create(frameSize_, CV_8UC1);
GpuMat fgmask = _fgmask.getGpuMat();
fgmask.setTo(Scalar::all(0), stream);
funcs[frame.depth()][frame.channels() - 1](frame, fgmask, colors_, weights_, nfeatures_, frameNum_,
learningRate_, updateBackgroundModel_, StreamAccessor::getStream(stream));
#if defined(HAVE_OPENCV_CUDAFILTERS) && defined(HAVE_OPENCV_CUDAARITHM)
// medianBlur
if (smoothingRadius_ > 0)
{
boxFilter_->apply(fgmask, buf_, stream);
const int minCount = (smoothingRadius_ * smoothingRadius_ + 1) / 2;
const double thresh = 255.0 * minCount / (smoothingRadius_ * smoothingRadius_);
cuda::threshold(buf_, fgmask, thresh, 255.0, THRESH_BINARY, stream);
}
#endif
// keep track of how many frames we have processed
++frameNum_;
}
void GMGImpl::getBackgroundImage(OutputArray backgroundImage) const
{
CV_UNUSED(backgroundImage);
CV_Error(Error::StsNotImplemented, "Not implemented");
}
void GMGImpl::initialize(Size frameSize, float min, float max)
{
using namespace cv::cuda::device::gmg;
CV_Assert( maxFeatures_ > 0 );
CV_Assert( learningRate_ >= 0.0f && learningRate_ <= 1.0f);
CV_Assert( numInitializationFrames_ >= 1);
CV_Assert( quantizationLevels_ >= 1 && quantizationLevels_ <= 255);
CV_Assert( backgroundPrior_ >= 0.0f && backgroundPrior_ <= 1.0f);
minVal_ = min;
maxVal_ = max;
CV_Assert( minVal_ < maxVal_ );
frameSize_ = frameSize;
frameNum_ = 0;
nfeatures_.create(frameSize_, CV_32SC1);
colors_.create(maxFeatures_ * frameSize_.height, frameSize_.width, CV_32SC1);
weights_.create(maxFeatures_ * frameSize_.height, frameSize_.width, CV_32FC1);
nfeatures_.setTo(Scalar::all(0));
#if defined(HAVE_OPENCV_CUDAFILTERS) && defined(HAVE_OPENCV_CUDAARITHM)
if (smoothingRadius_ > 0)
boxFilter_ = cuda::createBoxFilter(CV_8UC1, -1, Size(smoothingRadius_, smoothingRadius_));
#endif
loadConstants(frameSize_.width, frameSize_.height, minVal_, maxVal_,
quantizationLevels_, backgroundPrior_, decisionThreshold_, maxFeatures_, numInitializationFrames_);
}
}
Ptr<cuda::BackgroundSubtractorGMG> cv::cuda::createBackgroundSubtractorGMG(int initializationFrames, double decisionThreshold)
{
return makePtr<GMGImpl>(initializationFrames, decisionThreshold);
}
#endif
modules/cudalegacy/src/graphcuts.cpp
+283
View File
@@ -0,0 +1,283 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
// GraphCut has been removed in NPP 8.0
#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION >= 8000)
void cv::cuda::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::cuda::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
void cv::cuda::connectivityMask(const GpuMat&, GpuMat&, const cv::Scalar&, const cv::Scalar&, Stream&) { throw_no_cuda(); }
void cv::cuda::labelComponents(const GpuMat&, GpuMat&, int, Stream&) { throw_no_cuda(); }
#else /* !defined (HAVE_CUDA) */
namespace cv { namespace cuda { namespace device
{
namespace ccl
{
void labelComponents(const PtrStepSzb& edges, PtrStepSzi comps, int flags, cudaStream_t stream);
template<typename T>
void computeEdges(const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
}
}}}
static float4 scalarToCudaType(const cv::Scalar& in)
{
return make_float4((float)in[0], (float)in[1], (float)in[2], (float)in[3]);
}
void cv::cuda::connectivityMask(const GpuMat& image, GpuMat& mask, const cv::Scalar& lo, const cv::Scalar& hi, Stream& s)
{
CV_Assert(!image.empty());
int ch = image.channels();
CV_Assert(ch <= 4);
int depth = image.depth();
typedef void (*func_t)(const PtrStepSzb& image, PtrStepSzb edges, const float4& lo, const float4& hi, cudaStream_t stream);
static const func_t suppotLookup[8][4] =
{ // 1, 2, 3, 4
{ device::ccl::computeEdges<uchar>, 0, device::ccl::computeEdges<uchar3>, device::ccl::computeEdges<uchar4> },// CV_8U
{ 0, 0, 0, 0 },// CV_16U
{ device::ccl::computeEdges<ushort>, 0, device::ccl::computeEdges<ushort3>, device::ccl::computeEdges<ushort4> },// CV_8S
{ 0, 0, 0, 0 },// CV_16S
{ device::ccl::computeEdges<int>, 0, 0, 0 },// CV_32S
{ device::ccl::computeEdges<float>, 0, 0, 0 },// CV_32F
{ 0, 0, 0, 0 },// CV_64F
{ 0, 0, 0, 0 } // CV_16F
};
func_t f = suppotLookup[depth][ch - 1];
CV_Assert(f);
if (image.size() != mask.size() || mask.type() != CV_8UC1)
mask.create(image.size(), CV_8UC1);
cudaStream_t stream = StreamAccessor::getStream(s);
float4 culo = scalarToCudaType(lo), cuhi = scalarToCudaType(hi);
f(image, mask, culo, cuhi, stream);
}
void cv::cuda::labelComponents(const GpuMat& mask, GpuMat& components, int flags, Stream& s)
{
CV_Assert(!mask.empty() && mask.type() == CV_8U);
if (!deviceSupports(SHARED_ATOMICS))
CV_Error(cv::Error::StsNotImplemented, "The device doesn't support shared atomics and communicative synchronization!");
components.create(mask.size(), CV_32SC1);
cudaStream_t stream = StreamAccessor::getStream(s);
device::ccl::labelComponents(mask, components, flags, stream);
}
namespace
{
typedef NppStatus (*init_func_t)(NppiSize oSize, NppiGraphcutState** ppState, Npp8u* pDeviceMem);
class NppiGraphcutStateHandler
{
public:
NppiGraphcutStateHandler(NppiSize sznpp, Npp8u* pDeviceMem, const init_func_t func)
{
nppSafeCall( func(sznpp, &pState, pDeviceMem) );
}
~NppiGraphcutStateHandler()
{
nppSafeCall( nppiGraphcutFree(pState) );
}
operator NppiGraphcutState*()
{
return pState;
}
private:
NppiGraphcutState* pState;
};
}
void cv::cuda::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& bottom, GpuMat& labels, GpuMat& buf, Stream& s)
{
#if (CUDA_VERSION < 5000)
CV_Assert(terminals.type() == CV_32S);
#else
CV_Assert(terminals.type() == CV_32S || terminals.type() == CV_32F);
#endif
Size src_size = terminals.size();
CV_Assert(leftTransp.size() == Size(src_size.height, src_size.width));
CV_Assert(leftTransp.type() == terminals.type());
CV_Assert(rightTransp.size() == Size(src_size.height, src_size.width));
CV_Assert(rightTransp.type() == terminals.type());
CV_Assert(top.size() == src_size);
CV_Assert(top.type() == terminals.type());
CV_Assert(bottom.size() == src_size);
CV_Assert(bottom.type() == terminals.type());
labels.create(src_size, CV_8U);
NppiSize sznpp;
sznpp.width = src_size.width;
sznpp.height = src_size.height;
int bufsz;
nppSafeCall( nppiGraphcutGetSize(sznpp, &bufsz) );
ensureSizeIsEnough(1, bufsz, CV_8U, buf);
cudaStream_t stream = StreamAccessor::getStream(s);
NppStreamHandler h(stream);
NppiGraphcutStateHandler state(sznpp, buf.ptr<Npp8u>(), nppiGraphcutInitAlloc);
#if (CUDA_VERSION < 5000)
nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
#else
if (terminals.type() == CV_32S)
{
nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
}
else
{
nppSafeCall( nppiGraphcut_32f8u(terminals.ptr<Npp32f>(), leftTransp.ptr<Npp32f>(), rightTransp.ptr<Npp32f>(), top.ptr<Npp32f>(), bottom.ptr<Npp32f>(),
static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
}
#endif
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
void cv::cuda::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& topLeft, GpuMat& topRight,
GpuMat& bottom, GpuMat& bottomLeft, GpuMat& bottomRight, GpuMat& labels, GpuMat& buf, Stream& s)
{
#if (CUDA_VERSION < 5000)
CV_Assert(terminals.type() == CV_32S);
#else
CV_Assert(terminals.type() == CV_32S || terminals.type() == CV_32F);
#endif
Size src_size = terminals.size();
CV_Assert(leftTransp.size() == Size(src_size.height, src_size.width));
CV_Assert(leftTransp.type() == terminals.type());
CV_Assert(rightTransp.size() == Size(src_size.height, src_size.width));
CV_Assert(rightTransp.type() == terminals.type());
CV_Assert(top.size() == src_size);
CV_Assert(top.type() == terminals.type());
CV_Assert(topLeft.size() == src_size);
CV_Assert(topLeft.type() == terminals.type());
CV_Assert(topRight.size() == src_size);
CV_Assert(topRight.type() == terminals.type());
CV_Assert(bottom.size() == src_size);
CV_Assert(bottom.type() == terminals.type());
CV_Assert(bottomLeft.size() == src_size);
CV_Assert(bottomLeft.type() == terminals.type());
CV_Assert(bottomRight.size() == src_size);
CV_Assert(bottomRight.type() == terminals.type());
labels.create(src_size, CV_8U);
NppiSize sznpp;
sznpp.width = src_size.width;
sznpp.height = src_size.height;
int bufsz;
nppSafeCall( nppiGraphcut8GetSize(sznpp, &bufsz) );
ensureSizeIsEnough(1, bufsz, CV_8U, buf);
cudaStream_t stream = StreamAccessor::getStream(s);
NppStreamHandler h(stream);
NppiGraphcutStateHandler state(sznpp, buf.ptr<Npp8u>(), nppiGraphcut8InitAlloc);
#if (CUDA_VERSION < 5000)
nppSafeCall( nppiGraphcut8_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(),
top.ptr<Npp32s>(), topLeft.ptr<Npp32s>(), topRight.ptr<Npp32s>(),
bottom.ptr<Npp32s>(), bottomLeft.ptr<Npp32s>(), bottomRight.ptr<Npp32s>(),
static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
#else
if (terminals.type() == CV_32S)
{
nppSafeCall( nppiGraphcut8_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(),
top.ptr<Npp32s>(), topLeft.ptr<Npp32s>(), topRight.ptr<Npp32s>(),
bottom.ptr<Npp32s>(), bottomLeft.ptr<Npp32s>(), bottomRight.ptr<Npp32s>(),
static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
}
else
{
nppSafeCall( nppiGraphcut8_32f8u(terminals.ptr<Npp32f>(), leftTransp.ptr<Npp32f>(), rightTransp.ptr<Npp32f>(),
top.ptr<Npp32f>(), topLeft.ptr<Npp32f>(), topRight.ptr<Npp32f>(),
bottom.ptr<Npp32f>(), bottomLeft.ptr<Npp32f>(), bottomRight.ptr<Npp32f>(),
static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
}
#endif
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
#endif /* !defined (HAVE_CUDA) */
modules/cudalegacy/src/image_pyramid.cpp
+147
View File
@@ -0,0 +1,147 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined HAVE_CUDA || defined(CUDA_DISABLER)
Ptr<ImagePyramid> cv::cuda::createImagePyramid(InputArray, int, Stream&) { throw_no_cuda(); return Ptr<ImagePyramid>(); }
#else // HAVE_CUDA
namespace
{
class ImagePyramidImpl : public ImagePyramid
{
public:
ImagePyramidImpl(InputArray img, int nLayers, Stream& stream);
void getLayer(OutputArray outImg, Size outRoi, Stream& stream = Stream::Null()) const;
private:
GpuMat layer0_;
std::vector<GpuMat> pyramid_;
int nLayers_;
};
ImagePyramidImpl::ImagePyramidImpl(InputArray _img, int numLayers, Stream& stream)
{
GpuMat img = _img.getGpuMat();
CV_Assert( img.depth() <= CV_32F && img.channels() <= 4 );
img.copyTo(layer0_, stream);
Size szLastLayer = img.size();
nLayers_ = 1;
if (numLayers <= 0)
numLayers = 255; // it will cut-off when any of the dimensions goes 1
pyramid_.resize(numLayers);
for (int i = 0; i < numLayers - 1; ++i)
{
Size szCurLayer(szLastLayer.width / 2, szLastLayer.height / 2);
if (szCurLayer.width == 0 || szCurLayer.height == 0)
break;
ensureSizeIsEnough(szCurLayer, img.type(), pyramid_[i]);
nLayers_++;
const GpuMat& prevLayer = i == 0 ? layer0_ : pyramid_[i - 1];
cv::cuda::device::pyramid::downsampleX2(prevLayer, pyramid_[i], img.depth(), img.channels(), StreamAccessor::getStream(stream));
szLastLayer = szCurLayer;
}
}
void ImagePyramidImpl::getLayer(OutputArray _outImg, Size outRoi, Stream& stream) const
{
CV_Assert( outRoi.width <= layer0_.cols && outRoi.height <= layer0_.rows && outRoi.width > 0 && outRoi.height > 0 );
ensureSizeIsEnough(outRoi, layer0_.type(), _outImg);
GpuMat outImg = _outImg.getGpuMat();
if (outRoi.width == layer0_.cols && outRoi.height == layer0_.rows)
{
layer0_.copyTo(outImg, stream);
return;
}
float lastScale = 1.0f;
float curScale;
GpuMat lastLayer = layer0_;
GpuMat curLayer;
for (int i = 0; i < nLayers_ - 1; ++i)
{
curScale = lastScale * 0.5f;
curLayer = pyramid_[i];
if (outRoi.width == curLayer.cols && outRoi.height == curLayer.rows)
{
curLayer.copyTo(outImg, stream);
}
if (outRoi.width >= curLayer.cols && outRoi.height >= curLayer.rows)
break;
lastScale = curScale;
lastLayer = curLayer;
}
cv::cuda::device::pyramid::interpolateFrom1(lastLayer, outImg, outImg.depth(), outImg.channels(), StreamAccessor::getStream(stream));
}
}
Ptr<ImagePyramid> cv::cuda::createImagePyramid(InputArray img, int nLayers, Stream& stream)
{
return Ptr<ImagePyramid>(new ImagePyramidImpl(img, nLayers, stream));
}
#endif
modules/cudalegacy/src/interpolate_frames.cpp
+113
View File
@@ -0,0 +1,113 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined (HAVE_CUDA) || !defined (HAVE_OPENCV_CUDALEGACY) || defined (CUDA_DISABLER)
void cv::cuda::interpolateFrames(const GpuMat&, const GpuMat&, const GpuMat&, const GpuMat&, const GpuMat&, const GpuMat&, float, GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
#else
void cv::cuda::interpolateFrames(const GpuMat& frame0, const GpuMat& frame1, const GpuMat& fu, const GpuMat& fv, const GpuMat& bu, const GpuMat& bv,
float pos, GpuMat& newFrame, GpuMat& buf, Stream& s)
{
CV_Assert(frame0.type() == CV_32FC1);
CV_Assert(frame1.size() == frame0.size() && frame1.type() == frame0.type());
CV_Assert(fu.size() == frame0.size() && fu.type() == frame0.type());
CV_Assert(fv.size() == frame0.size() && fv.type() == frame0.type());
CV_Assert(bu.size() == frame0.size() && bu.type() == frame0.type());
CV_Assert(bv.size() == frame0.size() && bv.type() == frame0.type());
newFrame.create(frame0.size(), frame0.type());
buf.create(6 * frame0.rows, frame0.cols, CV_32FC1);
buf.setTo(Scalar::all(0));
// occlusion masks
GpuMat occ0 = buf.rowRange(0 * frame0.rows, 1 * frame0.rows);
GpuMat occ1 = buf.rowRange(1 * frame0.rows, 2 * frame0.rows);
// interpolated forward flow
GpuMat fui = buf.rowRange(2 * frame0.rows, 3 * frame0.rows);
GpuMat fvi = buf.rowRange(3 * frame0.rows, 4 * frame0.rows);
// interpolated backward flow
GpuMat bui = buf.rowRange(4 * frame0.rows, 5 * frame0.rows);
GpuMat bvi = buf.rowRange(5 * frame0.rows, 6 * frame0.rows);
size_t step = frame0.step;
CV_Assert(frame1.step == step && fu.step == step && fv.step == step && bu.step == step && bv.step == step && newFrame.step == step && buf.step == step);
cudaStream_t stream = StreamAccessor::getStream(s);
NppStStreamHandler h(stream);
NppStInterpolationState state;
state.size = NcvSize32u(frame0.cols, frame0.rows);
state.nStep = static_cast<Ncv32u>(step);
state.pSrcFrame0 = const_cast<Ncv32f*>(frame0.ptr<Ncv32f>());
state.pSrcFrame1 = const_cast<Ncv32f*>(frame1.ptr<Ncv32f>());
state.pFU = const_cast<Ncv32f*>(fu.ptr<Ncv32f>());
state.pFV = const_cast<Ncv32f*>(fv.ptr<Ncv32f>());
state.pBU = const_cast<Ncv32f*>(bu.ptr<Ncv32f>());
state.pBV = const_cast<Ncv32f*>(bv.ptr<Ncv32f>());
state.pos = pos;
state.pNewFrame = newFrame.ptr<Ncv32f>();
state.ppBuffers[0] = occ0.ptr<Ncv32f>();
state.ppBuffers[1] = occ1.ptr<Ncv32f>();
state.ppBuffers[2] = fui.ptr<Ncv32f>();
state.ppBuffers[3] = fvi.ptr<Ncv32f>();
state.ppBuffers[4] = bui.ptr<Ncv32f>();
state.ppBuffers[5] = bvi.ptr<Ncv32f>();
ncvSafeCall( nppiStInterpolateFrames(&state) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
#endif /* HAVE_CUDA */
modules/cudalegacy/src/needle_map.cpp
+100
View File
@@ -0,0 +1,100 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace cv;
using namespace cv::cuda;
#if !defined (HAVE_CUDA) || !defined(HAVE_OPENCV_CUDAIMGPROC) || defined (CUDA_DISABLER)
void cv::cuda::createOpticalFlowNeedleMap(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&) { throw_no_cuda(); }
#else
namespace cv { namespace cuda { namespace device
{
namespace optical_flow
{
void NeedleMapAverage_gpu(PtrStepSzf u, PtrStepSzf v, PtrStepSzf u_avg, PtrStepSzf v_avg);
void CreateOpticalFlowNeedleMap_gpu(PtrStepSzf u_avg, PtrStepSzf v_avg, float* vertex_buffer, float* color_data, float max_flow, float xscale, float yscale);
}
}}}
void cv::cuda::createOpticalFlowNeedleMap(const GpuMat& u, const GpuMat& v, GpuMat& vertex, GpuMat& colors)
{
using namespace cv::cuda::device::optical_flow;
CV_Assert(u.type() == CV_32FC1);
CV_Assert(v.type() == u.type() && v.size() == u.size());
const int NEEDLE_MAP_SCALE = 16;
const int x_needles = u.cols / NEEDLE_MAP_SCALE;
const int y_needles = u.rows / NEEDLE_MAP_SCALE;
GpuMat u_avg(y_needles, x_needles, CV_32FC1);
GpuMat v_avg(y_needles, x_needles, CV_32FC1);
NeedleMapAverage_gpu(u, v, u_avg, v_avg);
const int NUM_VERTS_PER_ARROW = 6;
const int num_arrows = x_needles * y_needles * NUM_VERTS_PER_ARROW;
vertex.create(1, num_arrows, CV_32FC3);
colors.create(1, num_arrows, CV_32FC3);
colors.setTo(Scalar::all(1.0));
double uMax, vMax;
cuda::minMax(u_avg, 0, &uMax);
cuda::minMax(v_avg, 0, &vMax);
float max_flow = static_cast<float>(std::sqrt(uMax * uMax + vMax * vMax));
CreateOpticalFlowNeedleMap_gpu(u_avg, v_avg, vertex.ptr<float>(), colors.ptr<float>(), max_flow, 1.0f / u.cols, 1.0f / u.rows);
cuda::cvtColor(colors, colors, COLOR_HSV2RGB);
}
#endif /* HAVE_CUDA */
modules/cudalegacy/src/precomp.hpp
+92
View File
@@ -0,0 +1,92 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_PRECOMP_H__
#define __OPENCV_PRECOMP_H__
#include <limits>
#include <iostream>
#include <algorithm>
#if defined(__OPENCV_BUILD) && defined(__clang__)
#pragma clang diagnostic ignored "-Winconsistent-missing-override"
#endif
#if defined(__OPENCV_BUILD) && defined(__GNUC__) && __GNUC__ >= 5
#pragma GCC diagnostic ignored "-Wsuggest-override"
#endif
#include "opencv2/geometry.hpp"
#include "opencv2/cudalegacy.hpp"
#include "opencv2/core/utility.hpp"
#include "opencv2/opencv_modules.hpp"
#ifdef HAVE_OPENCV_OBJDETECT
# include "opencv2/objdetect.hpp"
#endif
#ifdef HAVE_OPENCV_XOBJDETECT
# include "opencv2/xobjdetect.hpp"
#endif
#ifdef HAVE_OPENCV_3D
# include "opencv2/geometry.hpp"
#endif
#ifdef HAVE_OPENCV_CUDAARITHM
# include "opencv2/cudaarithm.hpp"
#endif
#ifdef HAVE_OPENCV_CUDAFILTERS
# include "opencv2/cudafilters.hpp"
#endif
#ifdef HAVE_OPENCV_CUDAIMGPROC
# include "opencv2/cudaimgproc.hpp"
#endif
#include "opencv2/core/private.cuda.hpp"
#if defined (HAVE_CUDA) && !defined (CUDA_DISABLER)
#include "opencv2/cudalegacy/private.hpp"
#endif
#endif /* __OPENCV_PRECOMP_H__ */
modules/cudalegacy/test/NCVAutoTestLister.hpp
+166
View File
@@ -0,0 +1,166 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncvautotestlister_hpp_
#define _ncvautotestlister_hpp_
#include <vector>
#include "NCVTest.hpp"
#include "main_test_nvidia.h"
class NCVAutoTestLister
{
public:
NCVAutoTestLister(std::string testSuiteName_, OutputLevel outputLevel_ = OutputLevelCompact, NcvBool bStopOnFirstFail_=false)
:
testSuiteName(testSuiteName_),
outputLevel(outputLevel_),
bStopOnFirstFail(bStopOnFirstFail_)
{
}
void add(INCVTest *test)
{
this->tests.push_back(test);
}
bool invoke()
{
Ncv32u nPassed = 0;
Ncv32u nFailed = 0;
Ncv32u nFailedMem = 0;
if (outputLevel == OutputLevelCompact)
{
printf("Test suite '%s' with %d tests\n",
testSuiteName.c_str(),
(int)(this->tests.size()));
}
for (Ncv32u i=0; i<this->tests.size(); i++)
{
INCVTest &curTest = *tests[i];
NCVTestReport curReport;
bool res = curTest.executeTest(curReport);
if (outputLevel == OutputLevelFull)
{
printf("Test %3i %16s; Consumed mem GPU = %8d, CPU = %8d; %s\n",
i,
curTest.getName().c_str(),
curReport.statsNums["MemGPU"],
curReport.statsNums["MemCPU"],
curReport.statsText["rcode"].c_str());
}
if (res)
{
nPassed++;
if (outputLevel == OutputLevelCompact)
{
printf(".");
}
}
else
{
if (!curReport.statsText["rcode"].compare("FAILED"))
{
nFailed++;
if (outputLevel == OutputLevelCompact)
{
printf("x");
}
if (bStopOnFirstFail)
{
break;
}
}
else
{
nFailedMem++;
if (outputLevel == OutputLevelCompact)
{
printf("m");
}
}
}
fflush(stdout);
}
if (outputLevel == OutputLevelCompact)
{
printf("\n");
}
if (outputLevel != OutputLevelNone)
{
printf("Test suite '%s' complete: %d total, %d passed, %d memory errors, %d failed\n\n",
testSuiteName.c_str(),
(int)(this->tests.size()),
nPassed,
nFailedMem,
nFailed);
}
bool passed = nFailed == 0 && nFailedMem == 0;
return passed;
}
~NCVAutoTestLister()
{
for (Ncv32u i=0; i<this->tests.size(); i++)
{
delete tests[i];
}
}
private:
std::string testSuiteName;
OutputLevel outputLevel;
NcvBool bStopOnFirstFail;
std::vector<INCVTest *> tests;
};
#endif // _ncvautotestlister_hpp_
modules/cudalegacy/test/NCVTest.hpp
+247
View File
@@ -0,0 +1,247 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncvtest_hpp_
#define _ncvtest_hpp_
#if defined _MSC_VER
# pragma warning( disable : 4201 4408 4100)
#endif
#include <string>
#include <vector>
#include <map>
#include <memory>
#include <algorithm>
#include <fstream>
#include <cuda_runtime.h>
#include "opencv2/cudalegacy.hpp"
struct NCVTestReport
{
std::map<std::string, Ncv32u> statsNums;
std::map<std::string, std::string> statsText;
};
class INCVTest
{
public:
virtual bool executeTest(NCVTestReport &report) = 0;
virtual std::string getName() const = 0;
virtual ~INCVTest(){}
};
class NCVTestProvider : public INCVTest
{
public:
NCVTestProvider(std::string testName_)
:
testName(testName_)
{
int devId;
ncvAssertPrintReturn(cudaSuccess == cudaGetDevice(&devId), "Error returned from cudaGetDevice", );
ncvAssertPrintReturn(cudaSuccess == cudaGetDeviceProperties(&this->devProp, devId), "Error returned from cudaGetDeviceProperties", );
}
virtual bool init() = 0;
virtual bool process() = 0;
virtual bool deinit() = 0;
virtual bool toString(std::ofstream &strOut) = 0;
virtual std::string getName() const
{
return this->testName;
}
virtual ~NCVTestProvider()
{
deinitMemory();
}
virtual bool executeTest(NCVTestReport &report)
{
bool res;
report.statsText["rcode"] = "FAILED";
res = initMemory(report);
if (!res)
{
dumpToFile(report);
deinitMemory();
return false;
}
res = init();
if (!res)
{
dumpToFile(report);
deinit();
deinitMemory();
return false;
}
res = process();
if (!res)
{
dumpToFile(report);
deinit();
deinitMemory();
return false;
}
res = deinit();
if (!res)
{
dumpToFile(report);
deinitMemory();
return false;
}
deinitMemory();
report.statsText["rcode"] = "Passed";
return true;
}
protected:
cudaDeviceProp devProp;
std::unique_ptr<INCVMemAllocator> allocatorGPU;
std::unique_ptr<INCVMemAllocator> allocatorCPU;
private:
std::string testName;
bool initMemory(NCVTestReport &report)
{
this->allocatorGPU.reset(new NCVMemStackAllocator(static_cast<Ncv32u>(devProp.textureAlignment)));
this->allocatorCPU.reset(new NCVMemStackAllocator(static_cast<Ncv32u>(devProp.textureAlignment)));
if (!this->allocatorGPU.get()->isInitialized() ||
!this->allocatorCPU.get()->isInitialized())
{
report.statsText["rcode"] = "Memory FAILED";
return false;
}
if (!this->process())
{
report.statsText["rcode"] = "Memory FAILED";
return false;
}
Ncv32u maxGPUsize = (Ncv32u)this->allocatorGPU.get()->maxSize();
Ncv32u maxCPUsize = (Ncv32u)this->allocatorCPU.get()->maxSize();
report.statsNums["MemGPU"] = maxGPUsize;
report.statsNums["MemCPU"] = maxCPUsize;
this->allocatorGPU.reset(new NCVMemStackAllocator(NCVMemoryTypeDevice, maxGPUsize, static_cast<Ncv32u>(devProp.textureAlignment)));
this->allocatorCPU.reset(new NCVMemStackAllocator(NCVMemoryTypeHostPinned, maxCPUsize, static_cast<Ncv32u>(devProp.textureAlignment)));
if (!this->allocatorGPU.get()->isInitialized() ||
!this->allocatorCPU.get()->isInitialized())
{
report.statsText["rcode"] = "Memory FAILED";
return false;
}
return true;
}
void deinitMemory()
{
this->allocatorGPU.reset();
this->allocatorCPU.reset();
}
void dumpToFile(NCVTestReport &report)
{
bool bReasonMem = (0 == report.statsText["rcode"].compare("Memory FAILED"));
std::string fname = "TestDump_";
fname += (bReasonMem ? "m_" : "") + this->testName + ".log";
std::ofstream stream(fname.c_str(), std::ios::trunc | std::ios::out);
if (!stream.is_open()) return;
stream << "NCV Test Failure Log: " << this->testName << std::endl;
stream << "====================================================" << std::endl << std::endl;
stream << "Test initialization report: " << std::endl;
for (std::map<std::string,std::string>::iterator it=report.statsText.begin();
it != report.statsText.end(); ++it)
{
stream << it->first << "=" << it->second << std::endl;
}
for (std::map<std::string,Ncv32u>::iterator it=report.statsNums.begin();
it != report.statsNums.end(); ++it)
{
stream << it->first << "=" << it->second << std::endl;
}
stream << std::endl;
stream << "Test initialization parameters: " << std::endl;
bool bSerializeRes = false;
try
{
bSerializeRes = this->toString(stream);
}
catch (...)
{
}
if (!bSerializeRes)
{
stream << "Couldn't retrieve object dump" << std::endl;
}
stream.flush();
}
};
#endif // _ncvtest_hpp_
modules/cudalegacy/test/NCVTestSourceProvider.hpp
+193
View File
@@ -0,0 +1,193 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _ncvtestsourceprovider_hpp_
#define _ncvtestsourceprovider_hpp_
#include <memory>
#include "opencv2/highgui.hpp"
#include "opencv2/cudalegacy.hpp"
template <class T>
class NCVTestSourceProvider
{
public:
NCVTestSourceProvider(Ncv32u seed, T rangeLow, T rangeHigh, Ncv32u maxWidth, Ncv32u maxHeight)
:
bInit(false)
{
ncvAssertPrintReturn(rangeLow < rangeHigh, "NCVTestSourceProvider ctor:: Invalid range", );
int devId;
cudaDeviceProp devProp;
ncvAssertPrintReturn(cudaSuccess == cudaGetDevice(&devId), "Error returned from cudaGetDevice", );
ncvAssertPrintReturn(cudaSuccess == cudaGetDeviceProperties(&devProp, devId), "Error returned from cudaGetDeviceProperties", );
//Ncv32u maxWpitch = alignUp(maxWidth * sizeof(T), devProp.textureAlignment);
allocatorCPU.reset(new NCVMemNativeAllocator(NCVMemoryTypeHostPinned, static_cast<Ncv32u>(devProp.textureAlignment)));
data.reset(new NCVMatrixAlloc<T>(*this->allocatorCPU.get(), maxWidth, maxHeight));
ncvAssertPrintReturn(data.get()->isMemAllocated(), "NCVTestSourceProvider ctor:: Matrix not allocated", );
this->dataWidth = maxWidth;
this->dataHeight = maxHeight;
srand(seed);
for (Ncv32u i=0; i<maxHeight; i++)
{
for (Ncv32u j=0; j<data.get()->stride(); j++)
{
data.get()->ptr()[i * data.get()->stride() + j] =
(T)(((1.0 * rand()) / RAND_MAX) * (rangeHigh - rangeLow) + rangeLow);
}
}
this->bInit = true;
}
NCVTestSourceProvider(std::string pgmFilename)
:
bInit(false)
{
ncvAssertPrintReturn(sizeof(T) == 1, "NCVTestSourceProvider ctor:: PGM constructor complies only with 8bit types", );
cv::Mat image = cv::imread(pgmFilename);
ncvAssertPrintReturn(!image.empty(), "NCVTestSourceProvider ctor:: PGM file error", );
int devId;
cudaDeviceProp devProp;
ncvAssertPrintReturn(cudaSuccess == cudaGetDevice(&devId), "Error returned from cudaGetDevice", );
ncvAssertPrintReturn(cudaSuccess == cudaGetDeviceProperties(&devProp, devId), "Error returned from cudaGetDeviceProperties", );
allocatorCPU.reset(new NCVMemNativeAllocator(NCVMemoryTypeHostPinned, static_cast<Ncv32u>(devProp.textureAlignment)));
data.reset(new NCVMatrixAlloc<T>(*this->allocatorCPU.get(), image.cols, image.rows));
ncvAssertPrintReturn(data.get()->isMemAllocated(), "NCVTestSourceProvider ctor:: Matrix not allocated", );
this->dataWidth = image.cols;
this->dataHeight = image.rows;
cv::Mat hdr(image.size(), CV_8UC1, data.get()->ptr(), data.get()->pitch());
image.copyTo(hdr);
this->bInit = true;
}
NcvBool fill(NCVMatrix<T> &dst)
{
ncvAssertReturn(this->isInit() &&
dst.memType() == allocatorCPU.get()->memType(), false);
if (dst.width() == 0 || dst.height() == 0)
{
return true;
}
for (Ncv32u i=0; i<dst.height(); i++)
{
Ncv32u srcLine = i % this->dataHeight;
Ncv32u srcFullChunks = dst.width() / this->dataWidth;
for (Ncv32u j=0; j<srcFullChunks; j++)
{
memcpy(dst.ptr() + i * dst.stride() + j * this->dataWidth,
this->data.get()->ptr() + this->data.get()->stride() * srcLine,
this->dataWidth * sizeof(T));
}
Ncv32u srcLastChunk = dst.width() % this->dataWidth;
memcpy(dst.ptr() + i * dst.stride() + srcFullChunks * this->dataWidth,
this->data.get()->ptr() + this->data.get()->stride() * srcLine,
srcLastChunk * sizeof(T));
}
return true;
}
NcvBool fill(NCVVector<T> &dst)
{
ncvAssertReturn(this->isInit() &&
dst.memType() == allocatorCPU.get()->memType(), false);
if (dst.length() == 0)
{
return true;
}
Ncv32u srcLen = this->dataWidth * this->dataHeight;
Ncv32u srcFullChunks = (Ncv32u)dst.length() / srcLen;
for (Ncv32u j=0; j<srcFullChunks; j++)
{
memcpy(dst.ptr() + j * srcLen, this->data.get()->ptr(), srcLen * sizeof(T));
}
Ncv32u srcLastChunk = dst.length() % srcLen;
memcpy(dst.ptr() + srcFullChunks * srcLen, this->data.get()->ptr(), srcLastChunk * sizeof(T));
return true;
}
~NCVTestSourceProvider()
{
data.reset();
allocatorCPU.reset();
}
private:
NcvBool isInit(void)
{
return this->bInit;
}
NcvBool bInit;
std::unique_ptr< INCVMemAllocator > allocatorCPU;
std::unique_ptr< NCVMatrixAlloc<T> > data;
Ncv32u dataWidth;
Ncv32u dataHeight;
};
#endif // _ncvtestsourceprovider_hpp_
modules/cudalegacy/test/TestCompact.cpp
+159
View File
@@ -0,0 +1,159 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
TestCompact::TestCompact(std::string testName_, NCVTestSourceProvider<Ncv32u> &src_,
Ncv32u length_, Ncv32u badElem_, Ncv32u badElemPercentage_)
:
NCVTestProvider(testName_),
src(src_),
length(length_),
badElem(badElem_),
badElemPercentage(badElemPercentage_ > 100 ? 100 : badElemPercentage_)
{
}
bool TestCompact::toString(std::ofstream &strOut)
{
strOut << "length=" << length << std::endl;
strOut << "badElem=" << badElem << std::endl;
strOut << "badElemPercentage=" << badElemPercentage << std::endl;
return true;
}
bool TestCompact::init()
{
return true;
}
bool TestCompact::process()
{
NCVStatus ncvStat;
bool rcode = false;
NCVVectorAlloc<Ncv32u> h_vecSrc(*this->allocatorCPU.get(), this->length);
ncvAssertReturn(h_vecSrc.isMemAllocated(), false);
NCVVectorAlloc<Ncv32u> d_vecSrc(*this->allocatorGPU.get(), this->length);
ncvAssertReturn(d_vecSrc.isMemAllocated(), false);
NCVVectorAlloc<Ncv32u> h_vecDst(*this->allocatorCPU.get(), this->length);
ncvAssertReturn(h_vecDst.isMemAllocated(), false);
NCVVectorAlloc<Ncv32u> d_vecDst(*this->allocatorGPU.get(), this->length);
ncvAssertReturn(d_vecDst.isMemAllocated(), false);
NCVVectorAlloc<Ncv32u> h_vecDst_d(*this->allocatorCPU.get(), this->length);
ncvAssertReturn(h_vecDst_d.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_vecSrc), false);
for (Ncv32u i=0; i<this->length; i++)
{
Ncv32u tmp = (h_vecSrc.ptr()[i]) & 0xFF;
tmp = tmp * 99 / 255;
if (tmp < this->badElemPercentage)
{
h_vecSrc.ptr()[i] = this->badElem;
}
}
NCV_SKIP_COND_END
NCVVectorAlloc<Ncv32u> h_dstLen(*this->allocatorCPU.get(), 1);
ncvAssertReturn(h_dstLen.isMemAllocated(), false);
Ncv32u bufSize;
ncvStat = nppsStCompactGetSize_32u(this->length, &bufSize, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
NCVVectorAlloc<Ncv8u> d_tmpBuf(*this->allocatorGPU.get(), bufSize);
ncvAssertReturn(d_tmpBuf.isMemAllocated(), false);
Ncv32u h_outElemNum_h = 0;
NCV_SKIP_COND_BEGIN
ncvStat = h_vecSrc.copySolid(d_vecSrc, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppsStCompact_32u(d_vecSrc.ptr(), this->length,
d_vecDst.ptr(), h_dstLen.ptr(), this->badElem,
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = d_vecDst.copySolid(h_vecDst_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppsStCompact_32u_host(h_vecSrc.ptr(), this->length, h_vecDst.ptr(), &h_outElemNum_h, this->badElem);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
if (h_dstLen.ptr()[0] != h_outElemNum_h)
{
bLoopVirgin = false;
}
else
{
for (Ncv32u i=0; bLoopVirgin && i < h_outElemNum_h; i++)
{
if (h_vecDst.ptr()[i] != h_vecDst_d.ptr()[i])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestCompact::deinit()
{
return true;
}
modules/cudalegacy/test/TestCompact.h
+73
View File
@@ -0,0 +1,73 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testhypothesescompact_h_
#define _testhypothesescompact_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestCompact : public NCVTestProvider
{
public:
TestCompact(std::string testName, NCVTestSourceProvider<Ncv32u> &src,
Ncv32u length, Ncv32u badElem, Ncv32u badElemPercentage);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestCompact(const TestCompact&);
TestCompact& operator=(const TestCompact&);
NCVTestSourceProvider<Ncv32u> &src;
Ncv32u length;
Ncv32u badElem;
Ncv32u badElemPercentage;
};
#endif // _testhypothesescompact_h_
modules/cudalegacy/test/TestDrawRects.cpp
+194
View File
@@ -0,0 +1,194 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
template <class T>
TestDrawRects<T>::TestDrawRects(std::string testName_, NCVTestSourceProvider<T> &src_,
NCVTestSourceProvider<Ncv32u> &src32u_,
Ncv32u width_, Ncv32u height_, Ncv32u numRects_, T color_)
:
NCVTestProvider(testName_),
src(src_),
src32u(src32u_),
width(width_),
height(height_),
numRects(numRects_),
color(color_)
{
}
template <class T>
bool TestDrawRects<T>::toString(std::ofstream &strOut)
{
strOut << "sizeof(T)=" << sizeof(T) << std::endl;
strOut << "width=" << width << std::endl;
strOut << "height=" << height << std::endl;
strOut << "numRects=" << numRects << std::endl;
strOut << "color=" << color << std::endl;
return true;
}
template <class T>
bool TestDrawRects<T>::init()
{
return true;
}
template <class T>
bool TestDrawRects<T>::process()
{
NCVStatus ncvStat;
bool rcode = false;
NCVMatrixAlloc<T> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img_d(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img_d.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> d_rects(*this->allocatorGPU.get(), this->numRects);
ncvAssertReturn(d_rects.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> h_rects(*this->allocatorCPU.get(), this->numRects);
ncvAssertReturn(h_rects.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
//fill vector of rectangles with random rects covering the input
NCVVectorReuse<Ncv32u> h_rects_as32u(h_rects.getSegment());
ncvAssertReturn(h_rects_as32u.isMemReused(), false);
ncvAssertReturn(this->src32u.fill(h_rects_as32u), false);
for (Ncv32u i=0; i<this->numRects; i++)
{
h_rects.ptr()[i].x = (Ncv32u)(((1.0 * h_rects.ptr()[i].x) / RAND_MAX) * (this->width-2));
h_rects.ptr()[i].y = (Ncv32u)(((1.0 * h_rects.ptr()[i].y) / RAND_MAX) * (this->height-2));
h_rects.ptr()[i].width = (Ncv32u)(((1.0 * h_rects.ptr()[i].width) / RAND_MAX) * (this->width+10 - h_rects.ptr()[i].x));
h_rects.ptr()[i].height = (Ncv32u)(((1.0 * h_rects.ptr()[i].height) / RAND_MAX) * (this->height+10 - h_rects.ptr()[i].y));
}
ncvStat = h_rects.copySolid(d_rects, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = ncvDrawRects_32u_device((Ncv32u *)d_img.ptr(), d_img.stride(), this->width, this->height,
(NcvRect32u *)d_rects.ptr(), this->numRects, this->color, 0);
}
else if (sizeof(T) == sizeof(Ncv8u))
{
ncvStat = ncvDrawRects_8u_device((Ncv8u *)d_img.ptr(), d_img.stride(), this->width, this->height,
(NcvRect32u *)d_rects.ptr(), this->numRects, (Ncv8u)this->color, 0);
}
else
{
ncvAssertPrintReturn(false, "Incorrect drawrects test instance", false);
}
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
ncvStat = d_img.copySolid(h_img_d, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = ncvDrawRects_32u_host((Ncv32u *)h_img.ptr(), h_img.stride(), this->width, this->height,
(NcvRect32u *)h_rects.ptr(), this->numRects, this->color);
}
else if (sizeof(T) == sizeof(Ncv8u))
{
ncvStat = ncvDrawRects_8u_host((Ncv8u *)h_img.ptr(), h_img.stride(), this->width, this->height,
(NcvRect32u *)h_rects.ptr(), this->numRects, (Ncv8u)this->color);
}
else
{
ncvAssertPrintReturn(false, "Incorrect drawrects test instance", false);
}
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
//const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < h_img.height(); i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_img.width(); j++)
{
if (h_img.ptr()[h_img.stride()*i+j] != h_img_d.ptr()[h_img_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
template <class T>
bool TestDrawRects<T>::deinit()
{
return true;
}
template class TestDrawRects<Ncv8u>;
template class TestDrawRects<Ncv32u>;
modules/cudalegacy/test/TestDrawRects.h
+76
View File
@@ -0,0 +1,76 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testdrawrects_h_
#define _testdrawrects_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
template <class T>
class TestDrawRects : public NCVTestProvider
{
public:
TestDrawRects(std::string testName, NCVTestSourceProvider<T> &src, NCVTestSourceProvider<Ncv32u> &src32u,
Ncv32u width, Ncv32u height, Ncv32u numRects, T color);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestDrawRects(const TestDrawRects&);
TestDrawRects& operator=(const TestDrawRects&);
NCVTestSourceProvider<T> &src;
NCVTestSourceProvider<Ncv32u> &src32u;
Ncv32u width;
Ncv32u height;
Ncv32u numRects;
T color;
};
#endif // _testdrawrects_h_
modules/cudalegacy/test/TestHaarCascadeApplication.cpp
+335
View File
@@ -0,0 +1,335 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
namespace
{
// http://www.christian-seiler.de/projekte/fpmath/
class FpuControl
{
public:
FpuControl();
~FpuControl();
private:
#if defined(__GNUC__) && !defined(__APPLE__) && !defined(__arm__) && !defined(__aarch64__) && !defined(__powerpc64__)
fpu_control_t fpu_oldcw, fpu_cw;
#elif defined(_WIN32) && !defined(_WIN64)
unsigned int fpu_oldcw, fpu_cw;
#endif
};
FpuControl::FpuControl()
{
#if defined(__GNUC__) && !defined(__APPLE__) && !defined(__arm__) && !defined(__aarch64__) && !defined(__powerpc64__)
_FPU_GETCW(fpu_oldcw);
fpu_cw = (fpu_oldcw & ~_FPU_EXTENDED & ~_FPU_DOUBLE & ~_FPU_SINGLE) | _FPU_SINGLE;
_FPU_SETCW(fpu_cw);
#elif defined(_WIN32) && !defined(_WIN64)
_controlfp_s(&fpu_cw, 0, 0);
fpu_oldcw = fpu_cw;
_controlfp_s(&fpu_cw, _PC_24, _MCW_PC);
#endif
}
FpuControl::~FpuControl()
{
#if defined(__GNUC__) && !defined(__APPLE__) && !defined(__arm__) && !defined(__aarch64__) && !defined(__powerpc64__)
_FPU_SETCW(fpu_oldcw);
#elif defined(_WIN32) && !defined(_WIN64)
_controlfp_s(&fpu_cw, fpu_oldcw, _MCW_PC);
#endif
}
}
TestHaarCascadeApplication::TestHaarCascadeApplication(std::string testName_, NCVTestSourceProvider<Ncv8u> &src_,
std::string cascadeName_, Ncv32u width_, Ncv32u height_)
:
NCVTestProvider(testName_),
src(src_),
cascadeName(cascadeName_),
width(width_),
height(height_)
{
}
bool TestHaarCascadeApplication::toString(std::ofstream &strOut)
{
strOut << "cascadeName=" << cascadeName << std::endl;
strOut << "width=" << width << std::endl;
strOut << "height=" << height << std::endl;
return true;
}
bool TestHaarCascadeApplication::init()
{
return true;
}
bool TestHaarCascadeApplication::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32u numStages, numNodes, numFeatures;
ncvStat = ncvHaarGetClassifierSize(this->cascadeName, numStages, numNodes, numFeatures);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCVVectorAlloc<HaarStage64> h_HaarStages(*this->allocatorCPU.get(), numStages);
ncvAssertReturn(h_HaarStages.isMemAllocated(), false);
NCVVectorAlloc<HaarClassifierNode128> h_HaarNodes(*this->allocatorCPU.get(), numNodes);
ncvAssertReturn(h_HaarNodes.isMemAllocated(), false);
NCVVectorAlloc<HaarFeature64> h_HaarFeatures(*this->allocatorCPU.get(), numFeatures);
ncvAssertReturn(h_HaarFeatures.isMemAllocated(), false);
NCVVectorAlloc<HaarStage64> d_HaarStages(*this->allocatorGPU.get(), numStages);
ncvAssertReturn(d_HaarStages.isMemAllocated(), false);
NCVVectorAlloc<HaarClassifierNode128> d_HaarNodes(*this->allocatorGPU.get(), numNodes);
ncvAssertReturn(d_HaarNodes.isMemAllocated(), false);
NCVVectorAlloc<HaarFeature64> d_HaarFeatures(*this->allocatorGPU.get(), numFeatures);
ncvAssertReturn(d_HaarFeatures.isMemAllocated(), false);
HaarClassifierCascadeDescriptor haar;
haar.ClassifierSize.width = haar.ClassifierSize.height = 1;
haar.bNeedsTiltedII = false;
haar.NumClassifierRootNodes = numNodes;
haar.NumClassifierTotalNodes = numNodes;
haar.NumFeatures = numFeatures;
haar.NumStages = numStages;
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvStat = ncvHaarLoadFromFile_host(this->cascadeName, haar, h_HaarStages, h_HaarNodes, h_HaarFeatures);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertReturn(NCV_SUCCESS == h_HaarStages.copySolid(d_HaarStages, 0), false);
ncvAssertReturn(NCV_SUCCESS == h_HaarNodes.copySolid(d_HaarNodes, 0), false);
ncvAssertReturn(NCV_SUCCESS == h_HaarFeatures.copySolid(d_HaarFeatures, 0), false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
NCV_SKIP_COND_END
NcvSize32s srcRoi, srcIIRoi, searchRoi;
srcRoi.width = this->width;
srcRoi.height = this->height;
srcIIRoi.width = srcRoi.width + 1;
srcIIRoi.height = srcRoi.height + 1;
searchRoi.width = srcIIRoi.width - haar.ClassifierSize.width;
searchRoi.height = srcIIRoi.height - haar.ClassifierSize.height;
if (searchRoi.width <= 0 || searchRoi.height <= 0)
{
return false;
}
NcvSize32u searchRoiU(searchRoi.width, searchRoi.height);
NCVMatrixAlloc<Ncv8u> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv8u> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
Ncv32u integralWidth = this->width + 1;
Ncv32u integralHeight = this->height + 1;
NCVMatrixAlloc<Ncv32u> d_integralImage(*this->allocatorGPU.get(), integralWidth, integralHeight);
ncvAssertReturn(d_integralImage.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> d_sqIntegralImage(*this->allocatorGPU.get(), integralWidth, integralHeight);
ncvAssertReturn(d_sqIntegralImage.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> h_integralImage(*this->allocatorCPU.get(), integralWidth, integralHeight);
ncvAssertReturn(h_integralImage.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> h_sqIntegralImage(*this->allocatorCPU.get(), integralWidth, integralHeight);
ncvAssertReturn(h_sqIntegralImage.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> d_rectStdDev(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_rectStdDev.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> d_pixelMask(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_pixelMask.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> h_rectStdDev(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_rectStdDev.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> h_pixelMask(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_pixelMask.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> d_hypotheses(*this->allocatorGPU.get(), this->width * this->height);
ncvAssertReturn(d_hypotheses.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> h_hypotheses(*this->allocatorCPU.get(), this->width * this->height);
ncvAssertReturn(h_hypotheses.isMemAllocated(), false);
NCVStatus nppStat;
Ncv32u szTmpBufIntegral, szTmpBufSqIntegral;
nppStat = nppiStIntegralGetSize_8u32u(NcvSize32u(this->width, this->height), &szTmpBufIntegral, this->devProp);
ncvAssertReturn(nppStat == NPPST_SUCCESS, false);
nppStat = nppiStSqrIntegralGetSize_8u64u(NcvSize32u(this->width, this->height), &szTmpBufSqIntegral, this->devProp);
ncvAssertReturn(nppStat == NPPST_SUCCESS, false);
NCVVectorAlloc<Ncv8u> d_tmpIIbuf(*this->allocatorGPU.get(), std::max(szTmpBufIntegral, szTmpBufSqIntegral));
ncvAssertReturn(d_tmpIIbuf.isMemAllocated(), false);
Ncv32u detectionsOnThisScale_d = 0;
Ncv32u detectionsOnThisScale_h = 0;
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
nppStat = nppiStIntegral_8u32u_C1R(d_img.ptr(), d_img.pitch(),
d_integralImage.ptr(), d_integralImage.pitch(),
NcvSize32u(d_img.width(), d_img.height()),
d_tmpIIbuf.ptr(), szTmpBufIntegral, this->devProp);
ncvAssertReturn(nppStat == NPPST_SUCCESS, false);
nppStat = nppiStSqrIntegral_8u64u_C1R(d_img.ptr(), d_img.pitch(),
d_sqIntegralImage.ptr(), d_sqIntegralImage.pitch(),
NcvSize32u(d_img.width(), d_img.height()),
d_tmpIIbuf.ptr(), szTmpBufSqIntegral, this->devProp);
ncvAssertReturn(nppStat == NPPST_SUCCESS, false);
const NcvRect32u rect(
HAAR_STDDEV_BORDER,
HAAR_STDDEV_BORDER,
haar.ClassifierSize.width - 2*HAAR_STDDEV_BORDER,
haar.ClassifierSize.height - 2*HAAR_STDDEV_BORDER);
nppStat = nppiStRectStdDev_32f_C1R(
d_integralImage.ptr(), d_integralImage.pitch(),
d_sqIntegralImage.ptr(), d_sqIntegralImage.pitch(),
d_rectStdDev.ptr(), d_rectStdDev.pitch(),
NcvSize32u(searchRoi.width, searchRoi.height), rect,
1.0f, true);
ncvAssertReturn(nppStat == NPPST_SUCCESS, false);
ncvStat = d_integralImage.copySolid(h_integralImage, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvStat = d_rectStdDev.copySolid(h_rectStdDev, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
for (Ncv32u i=0; i<searchRoiU.height; i++)
{
for (Ncv32u j=0; j<h_pixelMask.stride(); j++)
{
if (j<searchRoiU.width)
{
h_pixelMask.ptr()[i*h_pixelMask.stride()+j] = (i << 16) | j;
}
else
{
h_pixelMask.ptr()[i*h_pixelMask.stride()+j] = OBJDET_MASK_ELEMENT_INVALID_32U;
}
}
}
ncvAssertReturn(cudaSuccess == cudaStreamSynchronize(0), false);
{
// calculations here
FpuControl fpu;
CV_UNUSED(fpu);
ncvStat = ncvApplyHaarClassifierCascade_host(
h_integralImage, h_rectStdDev, h_pixelMask,
detectionsOnThisScale_h,
haar, h_HaarStages, h_HaarNodes, h_HaarFeatures, false,
searchRoiU, 1, 1.0f);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
}
NCV_SKIP_COND_END
int devId;
ncvAssertCUDAReturn(cudaGetDevice(&devId), false);
cudaDeviceProp _devProp;
ncvAssertCUDAReturn(cudaGetDeviceProperties(&_devProp, devId), false);
ncvStat = ncvApplyHaarClassifierCascade_device(
d_integralImage, d_rectStdDev, d_pixelMask,
detectionsOnThisScale_d,
haar, h_HaarStages, d_HaarStages, d_HaarNodes, d_HaarFeatures, false,
searchRoiU, 1, 1.0f,
*this->allocatorGPU.get(), *this->allocatorCPU.get(),
_devProp, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCVMatrixAlloc<Ncv32u> h_pixelMask_d(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_pixelMask_d.isMemAllocated(), false);
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
ncvStat = d_pixelMask.copySolid(h_pixelMask_d, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
if (detectionsOnThisScale_d != detectionsOnThisScale_h)
{
bLoopVirgin = false;
}
else
{
std::sort(h_pixelMask_d.ptr(), h_pixelMask_d.ptr() + detectionsOnThisScale_d);
for (Ncv32u i=0; i<detectionsOnThisScale_d && bLoopVirgin; i++)
{
if (h_pixelMask.ptr()[i] != h_pixelMask_d.ptr()[i])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestHaarCascadeApplication::deinit()
{
return true;
}
modules/cudalegacy/test/TestHaarCascadeApplication.h
+73
View File
@@ -0,0 +1,73 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testhaarcascadeapplication_h_
#define _testhaarcascadeapplication_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestHaarCascadeApplication : public NCVTestProvider
{
public:
TestHaarCascadeApplication(std::string testName, NCVTestSourceProvider<Ncv8u> &src,
std::string cascadeName, Ncv32u width, Ncv32u height);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestHaarCascadeApplication(const TestHaarCascadeApplication&);
TestHaarCascadeApplication& operator=(const TestHaarCascadeApplication&);
NCVTestSourceProvider<Ncv8u> &src;
std::string cascadeName;
Ncv32u width;
Ncv32u height;
};
#endif // _testhaarcascadeapplication_h_
modules/cudalegacy/test/TestHaarCascadeLoader.cpp
+153
View File
@@ -0,0 +1,153 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
TestHaarCascadeLoader::TestHaarCascadeLoader(std::string testName_, std::string cascadeName_)
:
NCVTestProvider(testName_),
cascadeName(cascadeName_)
{
}
bool TestHaarCascadeLoader::toString(std::ofstream &strOut)
{
strOut << "cascadeName=" << cascadeName << std::endl;
return true;
}
bool TestHaarCascadeLoader::init()
{
return true;
}
bool TestHaarCascadeLoader::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32u numStages, numNodes, numFeatures;
Ncv32u numStages_2 = 0, numNodes_2 = 0, numFeatures_2 = 0;
ncvStat = ncvHaarGetClassifierSize(this->cascadeName, numStages, numNodes, numFeatures);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCVVectorAlloc<HaarStage64> h_HaarStages(*this->allocatorCPU.get(), numStages);
ncvAssertReturn(h_HaarStages.isMemAllocated(), false);
NCVVectorAlloc<HaarClassifierNode128> h_HaarNodes(*this->allocatorCPU.get(), numNodes);
ncvAssertReturn(h_HaarNodes.isMemAllocated(), false);
NCVVectorAlloc<HaarFeature64> h_HaarFeatures(*this->allocatorCPU.get(), numFeatures);
ncvAssertReturn(h_HaarFeatures.isMemAllocated(), false);
NCVVectorAlloc<HaarStage64> h_HaarStages_2(*this->allocatorCPU.get(), numStages);
ncvAssertReturn(h_HaarStages_2.isMemAllocated(), false);
NCVVectorAlloc<HaarClassifierNode128> h_HaarNodes_2(*this->allocatorCPU.get(), numNodes);
ncvAssertReturn(h_HaarNodes_2.isMemAllocated(), false);
NCVVectorAlloc<HaarFeature64> h_HaarFeatures_2(*this->allocatorCPU.get(), numFeatures);
ncvAssertReturn(h_HaarFeatures_2.isMemAllocated(), false);
HaarClassifierCascadeDescriptor haar;
HaarClassifierCascadeDescriptor haar_2;
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
const std::string testNvbinName = cv::tempfile("test.nvbin");
ncvStat = ncvHaarLoadFromFile_host(this->cascadeName, haar, h_HaarStages, h_HaarNodes, h_HaarFeatures);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvStat = ncvHaarStoreNVBIN_host(testNvbinName, haar, h_HaarStages, h_HaarNodes, h_HaarFeatures);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvStat = ncvHaarGetClassifierSize(testNvbinName, numStages_2, numNodes_2, numFeatures_2);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvStat = ncvHaarLoadFromFile_host(testNvbinName, haar_2, h_HaarStages_2, h_HaarNodes_2, h_HaarFeatures_2);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
if (
numStages_2 != numStages ||
numNodes_2 != numNodes ||
numFeatures_2 != numFeatures ||
haar.NumStages != haar_2.NumStages ||
haar.NumClassifierRootNodes != haar_2.NumClassifierRootNodes ||
haar.NumClassifierTotalNodes != haar_2.NumClassifierTotalNodes ||
haar.NumFeatures != haar_2.NumFeatures ||
haar.ClassifierSize.width != haar_2.ClassifierSize.width ||
haar.ClassifierSize.height != haar_2.ClassifierSize.height ||
haar.bNeedsTiltedII != haar_2.bNeedsTiltedII ||
haar.bHasStumpsOnly != haar_2.bHasStumpsOnly )
{
bLoopVirgin = false;
}
if (memcmp(h_HaarStages.ptr(), h_HaarStages_2.ptr(), haar.NumStages * sizeof(HaarStage64)) ||
memcmp(h_HaarNodes.ptr(), h_HaarNodes_2.ptr(), haar.NumClassifierTotalNodes * sizeof(HaarClassifierNode128)) ||
memcmp(h_HaarFeatures.ptr(), h_HaarFeatures_2.ptr(), haar.NumFeatures * sizeof(HaarFeature64)) )
{
bLoopVirgin = false;
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestHaarCascadeLoader::deinit()
{
return true;
}
modules/cudalegacy/test/TestHaarCascadeLoader.h
+66
View File
@@ -0,0 +1,66 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testhaarcascadeloader_h_
#define _testhaarcascadeloader_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestHaarCascadeLoader : public NCVTestProvider
{
public:
TestHaarCascadeLoader(std::string testName, std::string cascadeName);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
std::string cascadeName;
};
#endif // _testhaarcascadeloader_h_
modules/cudalegacy/test/TestHypothesesFilter.cpp
+206
View File
@@ -0,0 +1,206 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
TestHypothesesFilter::TestHypothesesFilter(std::string testName_, NCVTestSourceProvider<Ncv32u> &src_,
Ncv32u numDstRects_, Ncv32u minNeighbors_, Ncv32f eps_)
:
NCVTestProvider(testName_),
src(src_),
numDstRects(numDstRects_),
minNeighbors(minNeighbors_),
eps(eps_)
{
}
bool TestHypothesesFilter::toString(std::ofstream &strOut)
{
strOut << "numDstRects=" << numDstRects << std::endl;
strOut << "minNeighbors=" << minNeighbors << std::endl;
strOut << "eps=" << eps << std::endl;
return true;
}
bool TestHypothesesFilter::init()
{
this->canvasWidth = 4096;
this->canvasHeight = 4096;
return true;
}
bool compareRects(const NcvRect32u &r1, const NcvRect32u &r2, Ncv32f eps)
{
double delta = eps*(std::min(r1.width, r2.width) + std::min(r1.height, r2.height))*0.5;
return std::abs((Ncv32s)r1.x - (Ncv32s)r2.x) <= delta &&
std::abs((Ncv32s)r1.y - (Ncv32s)r2.y) <= delta &&
std::abs((Ncv32s)r1.x + (Ncv32s)r1.width - (Ncv32s)r2.x - (Ncv32s)r2.width) <= delta &&
std::abs((Ncv32s)r1.y + (Ncv32s)r1.height - (Ncv32s)r2.y - (Ncv32s)r2.height) <= delta;
}
inline bool operator < (const NcvRect32u &a, const NcvRect32u &b)
{
return a.x < b.x;
}
bool TestHypothesesFilter::process()
{
NCVStatus ncvStat;
bool rcode = false;
NCVVectorAlloc<Ncv32u> h_random32u(*this->allocatorCPU.get(), this->numDstRects * sizeof(NcvRect32u) / sizeof(Ncv32u));
ncvAssertReturn(h_random32u.isMemAllocated(), false);
Ncv32u srcSlotSize = 2 * this->minNeighbors + 1;
NCVVectorAlloc<NcvRect32u> h_vecSrc(*this->allocatorCPU.get(), this->numDstRects*srcSlotSize);
ncvAssertReturn(h_vecSrc.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> h_vecDst_groundTruth(*this->allocatorCPU.get(), this->numDstRects);
ncvAssertReturn(h_vecDst_groundTruth.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorCPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_random32u), false);
Ncv32u randCnt = 0;
Ncv64f randVal;
for (Ncv32u i=0; i<this->numDstRects; i++)
{
h_vecDst_groundTruth.ptr()[i].x = i * this->canvasWidth / this->numDstRects + this->canvasWidth / (this->numDstRects * 4);
h_vecDst_groundTruth.ptr()[i].y = i * this->canvasHeight / this->numDstRects + this->canvasHeight / (this->numDstRects * 4);
h_vecDst_groundTruth.ptr()[i].width = this->canvasWidth / (this->numDstRects * 2);
h_vecDst_groundTruth.ptr()[i].height = this->canvasHeight / (this->numDstRects * 2);
Ncv32u numNeighbors = this->minNeighbors + 1 + (Ncv32u)(((1.0 * h_random32u.ptr()[i]) * (this->minNeighbors + 1)) / 0xFFFFFFFF);
numNeighbors = (numNeighbors > srcSlotSize) ? srcSlotSize : numNeighbors;
//fill in strong hypotheses (2 * ((1.0 * randVal) / 0xFFFFFFFF) - 1)
for (Ncv32u j=0; j<numNeighbors; j++)
{
randVal = (1.0 * h_random32u.ptr()[randCnt++]) / 0xFFFFFFFF; randCnt = randCnt % h_random32u.length();
h_vecSrc.ptr()[srcSlotSize * i + j].x =
h_vecDst_groundTruth.ptr()[i].x +
(Ncv32s)(h_vecDst_groundTruth.ptr()[i].width * this->eps * (randVal - 0.5));
randVal = (1.0 * h_random32u.ptr()[randCnt++]) / 0xFFFFFFFF; randCnt = randCnt % h_random32u.length();
h_vecSrc.ptr()[srcSlotSize * i + j].y =
h_vecDst_groundTruth.ptr()[i].y +
(Ncv32s)(h_vecDst_groundTruth.ptr()[i].height * this->eps * (randVal - 0.5));
h_vecSrc.ptr()[srcSlotSize * i + j].width = h_vecDst_groundTruth.ptr()[i].width;
h_vecSrc.ptr()[srcSlotSize * i + j].height = h_vecDst_groundTruth.ptr()[i].height;
}
//generate weak hypotheses (to be removed in processing)
for (Ncv32u j=numNeighbors; j<srcSlotSize; j++)
{
randVal = (1.0 * h_random32u.ptr()[randCnt++]) / 0xFFFFFFFF; randCnt = randCnt % h_random32u.length();
h_vecSrc.ptr()[srcSlotSize * i + j].x =
this->canvasWidth + h_vecDst_groundTruth.ptr()[i].x +
(Ncv32s)(h_vecDst_groundTruth.ptr()[i].width * this->eps * (randVal - 0.5));
randVal = (1.0 * h_random32u.ptr()[randCnt++]) / 0xFFFFFFFF; randCnt = randCnt % h_random32u.length();
h_vecSrc.ptr()[srcSlotSize * i + j].y =
this->canvasHeight + h_vecDst_groundTruth.ptr()[i].y +
(Ncv32s)(h_vecDst_groundTruth.ptr()[i].height * this->eps * (randVal - 0.5));
h_vecSrc.ptr()[srcSlotSize * i + j].width = h_vecDst_groundTruth.ptr()[i].width;
h_vecSrc.ptr()[srcSlotSize * i + j].height = h_vecDst_groundTruth.ptr()[i].height;
}
}
//shuffle
for (Ncv32u i=0; i<this->numDstRects*srcSlotSize-1; i++)
{
Ncv32u randValLocal = h_random32u.ptr()[randCnt++]; randCnt = randCnt % h_random32u.length();
Ncv32u secondSwap = randValLocal % (this->numDstRects*srcSlotSize-1 - i);
NcvRect32u tmp = h_vecSrc.ptr()[i + secondSwap];
h_vecSrc.ptr()[i + secondSwap] = h_vecSrc.ptr()[i];
h_vecSrc.ptr()[i] = tmp;
}
NCV_SKIP_COND_END
Ncv32u numHypothesesSrc = static_cast<Ncv32u>(h_vecSrc.length());
NCV_SKIP_COND_BEGIN
ncvStat = ncvGroupRectangles_host(h_vecSrc, numHypothesesSrc, this->minNeighbors, this->eps, NULL);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
//verification
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
if (numHypothesesSrc != this->numDstRects)
{
bLoopVirgin = false;
}
else
{
std::vector<NcvRect32u> tmpRects(numHypothesesSrc);
memcpy(&tmpRects[0], h_vecSrc.ptr(), numHypothesesSrc * sizeof(NcvRect32u));
std::sort(tmpRects.begin(), tmpRects.end());
for (Ncv32u i=0; i<numHypothesesSrc && bLoopVirgin; i++)
{
if (!compareRects(tmpRects[i], h_vecDst_groundTruth.ptr()[i], this->eps))
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestHypothesesFilter::deinit()
{
return true;
}
modules/cudalegacy/test/TestHypothesesFilter.h
+76
View File
@@ -0,0 +1,76 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testhypothesesfilter_h_
#define _testhypothesesfilter_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestHypothesesFilter : public NCVTestProvider
{
public:
TestHypothesesFilter(std::string testName, NCVTestSourceProvider<Ncv32u> &src,
Ncv32u numDstRects, Ncv32u minNeighbors, Ncv32f eps);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestHypothesesFilter(const TestHypothesesFilter&);
TestHypothesesFilter& operator=(const TestHypothesesFilter&);
NCVTestSourceProvider<Ncv32u> &src;
Ncv32u numDstRects;
Ncv32u minNeighbors;
Ncv32f eps;
Ncv32u canvasWidth;
Ncv32u canvasHeight;
};
#endif // _testhypothesesfilter_h_
modules/cudalegacy/test/TestHypothesesGrow.cpp
+165
View File
@@ -0,0 +1,165 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
TestHypothesesGrow::TestHypothesesGrow(std::string testName_, NCVTestSourceProvider<Ncv32u> &src_,
Ncv32u rectWidth_, Ncv32u rectHeight_, Ncv32f rectScale_,
Ncv32u maxLenSrc_, Ncv32u lenSrc_, Ncv32u maxLenDst_, Ncv32u lenDst_)
:
NCVTestProvider(testName_),
src(src_),
rectWidth(rectWidth_),
rectHeight(rectHeight_),
rectScale(rectScale_),
maxLenSrc(maxLenSrc_),
lenSrc(lenSrc_),
maxLenDst(maxLenDst_),
lenDst(lenDst_)
{
}
bool TestHypothesesGrow::toString(std::ofstream &strOut)
{
strOut << "rectWidth=" << rectWidth << std::endl;
strOut << "rectHeight=" << rectHeight << std::endl;
strOut << "rectScale=" << rectScale << std::endl;
strOut << "maxLenSrc=" << maxLenSrc << std::endl;
strOut << "lenSrc=" << lenSrc << std::endl;
strOut << "maxLenDst=" << maxLenDst << std::endl;
strOut << "lenDst=" << lenDst << std::endl;
return true;
}
bool TestHypothesesGrow::init()
{
return true;
}
bool TestHypothesesGrow::process()
{
NCVStatus ncvStat;
bool rcode = false;
NCVVectorAlloc<Ncv32u> h_vecSrc(*this->allocatorCPU.get(), this->maxLenSrc);
ncvAssertReturn(h_vecSrc.isMemAllocated(), false);
NCVVectorAlloc<Ncv32u> d_vecSrc(*this->allocatorGPU.get(), this->maxLenSrc);
ncvAssertReturn(d_vecSrc.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> h_vecDst(*this->allocatorCPU.get(), this->maxLenDst);
ncvAssertReturn(h_vecDst.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> d_vecDst(*this->allocatorGPU.get(), this->maxLenDst);
ncvAssertReturn(d_vecDst.isMemAllocated(), false);
NCVVectorAlloc<NcvRect32u> h_vecDst_d(*this->allocatorCPU.get(), this->maxLenDst);
ncvAssertReturn(h_vecDst_d.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_vecSrc), false);
*h_vecDst.ptr() = {};
NCVVectorReuse<Ncv32u> h_vecDst_as32u(h_vecDst.getSegment(), lenDst * sizeof(NcvRect32u) / sizeof(Ncv32u));
ncvAssertReturn(h_vecDst_as32u.isMemReused(), false);
ncvAssertReturn(this->src.fill(h_vecDst_as32u), false);
memcpy(h_vecDst_d.ptr(), h_vecDst.ptr(), h_vecDst.length() * sizeof(NcvRect32u));
NCV_SKIP_COND_END
ncvStat = h_vecSrc.copySolid(d_vecSrc, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvStat = h_vecDst.copySolid(d_vecDst, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
Ncv32u h_outElemNum_d = 0;
Ncv32u h_outElemNum_h = 0;
NCV_SKIP_COND_BEGIN
h_outElemNum_d = this->lenDst;
ncvStat = ncvGrowDetectionsVector_device(d_vecSrc, this->lenSrc,
d_vecDst, h_outElemNum_d, this->maxLenDst,
this->rectWidth, this->rectHeight, this->rectScale, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvStat = d_vecDst.copySolid(h_vecDst_d, 0);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
ncvAssertCUDAReturn(cudaStreamSynchronize(0), false);
h_outElemNum_h = this->lenDst;
ncvStat = ncvGrowDetectionsVector_host(h_vecSrc, this->lenSrc,
h_vecDst, h_outElemNum_h, this->maxLenDst,
this->rectWidth, this->rectHeight, this->rectScale);
ncvAssertReturn(ncvStat == NCV_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
if (h_outElemNum_d != h_outElemNum_h)
{
bLoopVirgin = false;
}
else
{
if (memcmp(h_vecDst.ptr(), h_vecDst_d.ptr(), this->maxLenDst * sizeof(NcvRect32u)))
{
bLoopVirgin = false;
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestHypothesesGrow::deinit()
{
return true;
}
modules/cudalegacy/test/TestHypothesesGrow.h
+78
View File
@@ -0,0 +1,78 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testhypothesesgrow_h_
#define _testhypothesesgrow_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestHypothesesGrow : public NCVTestProvider
{
public:
TestHypothesesGrow(std::string testName, NCVTestSourceProvider<Ncv32u> &src,
Ncv32u rectWidth, Ncv32u rectHeight, Ncv32f rectScale,
Ncv32u maxLenSrc, Ncv32u lenSrc, Ncv32u maxLenDst, Ncv32u lenDst);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestHypothesesGrow(const TestHypothesesGrow&);
TestHypothesesGrow& operator=(const TestHypothesesGrow&);
NCVTestSourceProvider<Ncv32u> &src;
Ncv32u rectWidth;
Ncv32u rectHeight;
Ncv32f rectScale;
Ncv32u maxLenSrc;
Ncv32u lenSrc;
Ncv32u maxLenDst;
Ncv32u lenDst;
};
#endif // _testhypothesesgrow_h_
modules/cudalegacy/test/TestIntegralImage.cpp
+215
View File
@@ -0,0 +1,215 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
template <class T_in, class T_out>
TestIntegralImage<T_in, T_out>::TestIntegralImage(std::string testName_, NCVTestSourceProvider<T_in> &src_,
Ncv32u width_, Ncv32u height_)
:
NCVTestProvider(testName_),
src(src_),
width(width_),
height(height_)
{
}
template <class T_in, class T_out>
bool TestIntegralImage<T_in, T_out>::toString(std::ofstream &strOut)
{
strOut << "sizeof(T_in)=" << sizeof(T_in) << std::endl;
strOut << "sizeof(T_out)=" << sizeof(T_out) << std::endl;
strOut << "width=" << width << std::endl;
strOut << "height=" << height << std::endl;
return true;
}
template <class T_in, class T_out>
bool TestIntegralImage<T_in, T_out>::init()
{
return true;
}
template <class T_in, class T_out>
bool TestIntegralImage<T_in, T_out>::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32u widthII = this->width + 1;
Ncv32u heightII = this->height + 1;
NCVMatrixAlloc<T_in> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T_in> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T_out> d_imgII(*this->allocatorGPU.get(), widthII, heightII);
ncvAssertReturn(d_imgII.isMemAllocated(), false);
NCVMatrixAlloc<T_out> h_imgII(*this->allocatorCPU.get(), widthII, heightII);
ncvAssertReturn(h_imgII.isMemAllocated(), false);
NCVMatrixAlloc<T_out> h_imgII_d(*this->allocatorCPU.get(), widthII, heightII);
ncvAssertReturn(h_imgII_d.isMemAllocated(), false);
Ncv32u bufSize;
if (sizeof(T_in) == sizeof(Ncv8u))
{
ncvStat = nppiStIntegralGetSize_8u32u(NcvSize32u(this->width, this->height), &bufSize, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
ncvStat = nppiStIntegralGetSize_32f32f(NcvSize32u(this->width, this->height), &bufSize, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
NCVVectorAlloc<Ncv8u> d_tmpBuf(*this->allocatorGPU.get(), bufSize);
ncvAssertReturn(d_tmpBuf.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
if (sizeof(T_in) == sizeof(Ncv8u))
{
ncvStat = nppiStIntegral_8u32u_C1R((Ncv8u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_imgII.ptr(), d_imgII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
ncvStat = nppiStIntegral_32f32f_C1R((Ncv32f *)d_img.ptr(), d_img.pitch(),
(Ncv32f *)d_imgII.ptr(), d_imgII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
ncvStat = d_imgII.copySolid(h_imgII_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
if (sizeof(T_in) == sizeof(Ncv8u))
{
ncvStat = nppiStIntegral_8u32u_C1R_host((Ncv8u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_imgII.ptr(), h_imgII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
ncvStat = nppiStIntegral_32f32f_C1R_host((Ncv32f *)h_img.ptr(), h_img.pitch(),
(Ncv32f *)h_imgII.ptr(), h_imgII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
for (Ncv32u i=0; bLoopVirgin && i < h_img.height() + 1; i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_img.width() + 1; j++)
{
if (sizeof(T_in) == sizeof(Ncv8u))
{
if (h_imgII.ptr()[h_imgII.stride()*i+j] != h_imgII_d.ptr()[h_imgII_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
else if (sizeof(T_in) == sizeof(Ncv32f))
{
if (fabsf((float)h_imgII.ptr()[h_imgII.stride()*i+j] - (float)h_imgII_d.ptr()[h_imgII_d.stride()*i+j]) > 0.01f)
{
bLoopVirgin = false;
}
}
else
{
ncvAssertPrintReturn(false, "Incorrect integral image test instance", false);
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
template <class T_in, class T_out>
bool TestIntegralImage<T_in, T_out>::deinit()
{
return true;
}
template class TestIntegralImage<Ncv8u, Ncv32u>;
template class TestIntegralImage<Ncv32f, Ncv32f>;
modules/cudalegacy/test/TestIntegralImage.h
+72
View File
@@ -0,0 +1,72 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testintegralimage_h_
#define _testintegralimage_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
template <class T_in, class T_out>
class TestIntegralImage : public NCVTestProvider
{
public:
TestIntegralImage(std::string testName, NCVTestSourceProvider<T_in> &src,
Ncv32u width, Ncv32u height);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestIntegralImage(const TestIntegralImage&);
TestIntegralImage& operator=(const TestIntegralImage&);
NCVTestSourceProvider<T_in> &src;
Ncv32u width;
Ncv32u height;
};
#endif // _testintegralimage_h_
modules/cudalegacy/test/TestIntegralImageSquared.cpp
+148
View File
@@ -0,0 +1,148 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
TestIntegralImageSquared::TestIntegralImageSquared(std::string testName_, NCVTestSourceProvider<Ncv8u> &src_,
Ncv32u width_, Ncv32u height_)
:
NCVTestProvider(testName_),
src(src_),
width(width_),
height(height_)
{
}
bool TestIntegralImageSquared::toString(std::ofstream &strOut)
{
strOut << "width=" << width << std::endl;
strOut << "height=" << height << std::endl;
return true;
}
bool TestIntegralImageSquared::init()
{
return true;
}
bool TestIntegralImageSquared::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32u widthSII = this->width + 1;
Ncv32u heightSII = this->height + 1;
NCVMatrixAlloc<Ncv8u> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv8u> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> d_imgSII(*this->allocatorGPU.get(), widthSII, heightSII);
ncvAssertReturn(d_imgSII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> h_imgSII(*this->allocatorCPU.get(), widthSII, heightSII);
ncvAssertReturn(h_imgSII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> h_imgSII_d(*this->allocatorCPU.get(), widthSII, heightSII);
ncvAssertReturn(h_imgSII_d.isMemAllocated(), false);
Ncv32u bufSize;
ncvStat = nppiStSqrIntegralGetSize_8u64u(NcvSize32u(this->width, this->height), &bufSize, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
NCVVectorAlloc<Ncv8u> d_tmpBuf(*this->allocatorGPU.get(), bufSize);
ncvAssertReturn(d_tmpBuf.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStSqrIntegral_8u64u_C1R(d_img.ptr(), d_img.pitch(),
d_imgSII.ptr(), d_imgSII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = d_imgSII.copySolid(h_imgSII_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStSqrIntegral_8u64u_C1R_host(h_img.ptr(), h_img.pitch(),
h_imgSII.ptr(), h_imgSII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
for (Ncv32u i=0; bLoopVirgin && i < h_img.height() + 1; i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_img.width() + 1; j++)
{
if (h_imgSII.ptr()[h_imgSII.stride()*i+j] != h_imgSII_d.ptr()[h_imgSII_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestIntegralImageSquared::deinit()
{
return true;
}
modules/cudalegacy/test/TestIntegralImageSquared.h
+71
View File
@@ -0,0 +1,71 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testintegralimagesquared_h_
#define _testintegralimagesquared_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestIntegralImageSquared : public NCVTestProvider
{
public:
TestIntegralImageSquared(std::string testName, NCVTestSourceProvider<Ncv8u> &src,
Ncv32u width, Ncv32u height);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestIntegralImageSquared(const TestIntegralImageSquared&);
TestIntegralImageSquared& operator=(const TestIntegralImageSquared&);
NCVTestSourceProvider<Ncv8u> &src;
Ncv32u width;
Ncv32u height;
};
#endif // _testintegralimagesquared_h_
modules/cudalegacy/test/TestRectStdDev.cpp
+209
View File
@@ -0,0 +1,209 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
TestRectStdDev::TestRectStdDev(std::string testName_, NCVTestSourceProvider<Ncv8u> &src_,
Ncv32u width_, Ncv32u height_, NcvRect32u rect_, Ncv32f scaleFactor_,
NcvBool bTextureCache_)
:
NCVTestProvider(testName_),
src(src_),
width(width_),
height(height_),
rect(rect_),
scaleFactor(scaleFactor_),
bTextureCache(bTextureCache_)
{
}
bool TestRectStdDev::toString(std::ofstream &strOut)
{
strOut << "width=" << width << std::endl;
strOut << "height=" << height << std::endl;
strOut << "rect=[" << rect.x << ", " << rect.y << ", " << rect.width << ", " << rect.height << "]\n";
strOut << "scaleFactor=" << scaleFactor << std::endl;
strOut << "bTextureCache=" << bTextureCache << std::endl;
return true;
}
bool TestRectStdDev::init()
{
return true;
}
bool TestRectStdDev::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32s _normWidth = (Ncv32s)this->width - this->rect.x - this->rect.width + 1;
Ncv32s _normHeight = (Ncv32s)this->height - this->rect.y - this->rect.height + 1;
if (_normWidth <= 0 || _normHeight <= 0)
{
return true;
}
Ncv32u normWidth = (Ncv32u)_normWidth;
Ncv32u normHeight = (Ncv32u)_normHeight;
NcvSize32u szNormRoi(normWidth, normHeight);
Ncv32u widthII = this->width + 1;
Ncv32u heightII = this->height + 1;
Ncv32u widthSII = this->width + 1;
Ncv32u heightSII = this->height + 1;
NCVMatrixAlloc<Ncv8u> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv8u> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> d_imgII(*this->allocatorGPU.get(), widthII, heightII);
ncvAssertReturn(d_imgII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32u> h_imgII(*this->allocatorCPU.get(), widthII, heightII);
ncvAssertReturn(h_imgII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> d_imgSII(*this->allocatorGPU.get(), widthSII, heightSII);
ncvAssertReturn(d_imgSII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv64u> h_imgSII(*this->allocatorCPU.get(), widthSII, heightSII);
ncvAssertReturn(h_imgSII.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> d_norm(*this->allocatorGPU.get(), normWidth, normHeight);
ncvAssertReturn(d_norm.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> h_norm(*this->allocatorCPU.get(), normWidth, normHeight);
ncvAssertReturn(h_norm.isMemAllocated(), false);
NCVMatrixAlloc<Ncv32f> h_norm_d(*this->allocatorCPU.get(), normWidth, normHeight);
ncvAssertReturn(h_norm_d.isMemAllocated(), false);
Ncv32u bufSizeII, bufSizeSII;
ncvStat = nppiStIntegralGetSize_8u32u(NcvSize32u(this->width, this->height), &bufSizeII, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
ncvStat = nppiStSqrIntegralGetSize_8u64u(NcvSize32u(this->width, this->height), &bufSizeSII, this->devProp);
ncvAssertReturn(NPPST_SUCCESS == ncvStat, false);
Ncv32u bufSize = bufSizeII > bufSizeSII ? bufSizeII : bufSizeSII;
NCVVectorAlloc<Ncv8u> d_tmpBuf(*this->allocatorGPU.get(), bufSize);
ncvAssertReturn(d_tmpBuf.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStIntegral_8u32u_C1R(d_img.ptr(), d_img.pitch(),
d_imgII.ptr(), d_imgII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStSqrIntegral_8u64u_C1R(d_img.ptr(), d_img.pitch(),
d_imgSII.ptr(), d_imgSII.pitch(),
NcvSize32u(this->width, this->height),
d_tmpBuf.ptr(), bufSize, this->devProp);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStRectStdDev_32f_C1R(d_imgII.ptr(), d_imgII.pitch(),
d_imgSII.ptr(), d_imgSII.pitch(),
d_norm.ptr(), d_norm.pitch(),
szNormRoi, this->rect,
this->scaleFactor,
this->bTextureCache);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = d_norm.copySolid(h_norm_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStIntegral_8u32u_C1R_host(h_img.ptr(), h_img.pitch(),
h_imgII.ptr(), h_imgII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStSqrIntegral_8u64u_C1R_host(h_img.ptr(), h_img.pitch(),
h_imgSII.ptr(), h_imgSII.pitch(),
NcvSize32u(this->width, this->height));
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
ncvStat = nppiStRectStdDev_32f_C1R_host(h_imgII.ptr(), h_imgII.pitch(),
h_imgSII.ptr(), h_imgSII.pitch(),
h_norm.ptr(), h_norm.pitch(),
szNormRoi, this->rect,
this->scaleFactor);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < h_norm.height(); i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_norm.width(); j++)
{
Ncv64f absErr = fabs(h_norm.ptr()[h_norm.stride()*i+j] - h_norm_d.ptr()[h_norm_d.stride()*i+j]);
Ncv64f relErr = absErr / h_norm.ptr()[h_norm.stride()*i+j];
if (relErr > relEPS)
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
bool TestRectStdDev::deinit()
{
return true;
}
modules/cudalegacy/test/TestRectStdDev.h
+76
View File
@@ -0,0 +1,76 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testrectstddev_h_
#define _testrectstddev_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
class TestRectStdDev : public NCVTestProvider
{
public:
TestRectStdDev(std::string testName, NCVTestSourceProvider<Ncv8u> &src,
Ncv32u width, Ncv32u height, NcvRect32u rect, Ncv32f scaleFactor,
NcvBool bTextureCache);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestRectStdDev(const TestRectStdDev&);
TestRectStdDev& operator=(const TestRectStdDev&);
NCVTestSourceProvider<Ncv8u> &src;
Ncv32u width;
Ncv32u height;
NcvRect32u rect;
Ncv32f scaleFactor;
NcvBool bTextureCache;
};
#endif // _testrectstddev_h_
modules/cudalegacy/test/TestResize.cpp
+190
View File
@@ -0,0 +1,190 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
template <class T>
TestResize<T>::TestResize(std::string testName_, NCVTestSourceProvider<T> &src_,
Ncv32u width_, Ncv32u height_, Ncv32u scaleFactor_, NcvBool bTextureCache_)
:
NCVTestProvider(testName_),
src(src_),
width(width_),
height(height_),
scaleFactor(scaleFactor_),
bTextureCache(bTextureCache_)
{
}
template <class T>
bool TestResize<T>::toString(std::ofstream &strOut)
{
strOut << "sizeof(T)=" << sizeof(T) << std::endl;
strOut << "width=" << width << std::endl;
strOut << "scaleFactor=" << scaleFactor << std::endl;
strOut << "bTextureCache=" << bTextureCache << std::endl;
return true;
}
template <class T>
bool TestResize<T>::init()
{
return true;
}
template <class T>
bool TestResize<T>::process()
{
NCVStatus ncvStat;
bool rcode = false;
Ncv32s smallWidth = this->width / this->scaleFactor;
Ncv32s smallHeight = this->height / this->scaleFactor;
if (smallWidth == 0 || smallHeight == 0)
{
return true;
}
NcvSize32u srcSize(this->width, this->height);
NCVMatrixAlloc<T> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T> d_small(*this->allocatorGPU.get(), smallWidth, smallHeight);
ncvAssertReturn(d_small.isMemAllocated(), false);
NCVMatrixAlloc<T> h_small(*this->allocatorCPU.get(), smallWidth, smallHeight);
ncvAssertReturn(h_small.isMemAllocated(), false);
NCVMatrixAlloc<T> h_small_d(*this->allocatorCPU.get(), smallWidth, smallHeight);
ncvAssertReturn(h_small_d.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
NCV_SKIP_COND_END
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStDecimate_32u_C1R((Ncv32u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStDecimate_64u_C1R((Ncv64u *)d_img.ptr(), d_img.pitch(),
(Ncv64u *)d_small.ptr(), d_small.pitch(),
srcSize, this->scaleFactor,
this->bTextureCache);
}
else
{
ncvAssertPrintReturn(false, "Incorrect downsample test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
ncvStat = d_small.copySolid(h_small_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStDecimate_32u_C1R_host((Ncv32u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStDecimate_64u_C1R_host((Ncv64u *)h_img.ptr(), h_img.pitch(),
(Ncv64u *)h_small.ptr(), h_small.pitch(),
srcSize, this->scaleFactor);
}
else
{
ncvAssertPrintReturn(false, "Incorrect downsample test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
//const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < h_small.height(); i++)
{
for (Ncv32u j=0; bLoopVirgin && j < h_small.width(); j++)
{
if (h_small.ptr()[h_small.stride()*i+j] != h_small_d.ptr()[h_small_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
template <class T>
bool TestResize<T>::deinit()
{
return true;
}
template class TestResize<Ncv32u>;
template class TestResize<Ncv64u>;
modules/cudalegacy/test/TestResize.h
+74
View File
@@ -0,0 +1,74 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testresize_h_
#define _testresize_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
template <class T>
class TestResize : public NCVTestProvider
{
public:
TestResize(std::string testName, NCVTestSourceProvider<T> &src,
Ncv32u width, Ncv32u height, Ncv32u scaleFactor, NcvBool bTextureCache);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestResize(const TestResize&);
TestResize& operator=(const TestResize&);
NCVTestSourceProvider<T> &src;
Ncv32u width;
Ncv32u height;
Ncv32u scaleFactor;
NcvBool bTextureCache;
};
#endif // _testresize_h_
modules/cudalegacy/test/TestTranspose.cpp
+177
View File
@@ -0,0 +1,177 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
template <class T>
TestTranspose<T>::TestTranspose(std::string testName_, NCVTestSourceProvider<T> &src_,
Ncv32u width_, Ncv32u height_)
:
NCVTestProvider(testName_),
src(src_),
width(width_),
height(height_)
{
}
template <class T>
bool TestTranspose<T>::toString(std::ofstream &strOut)
{
strOut << "sizeof(T)=" << sizeof(T) << std::endl;
strOut << "width=" << width << std::endl;
return true;
}
template <class T>
bool TestTranspose<T>::init()
{
return true;
}
template <class T>
bool TestTranspose<T>::process()
{
NCVStatus ncvStat;
bool rcode = false;
NcvSize32u srcSize(this->width, this->height);
NCVMatrixAlloc<T> d_img(*this->allocatorGPU.get(), this->width, this->height);
ncvAssertReturn(d_img.isMemAllocated(), false);
NCVMatrixAlloc<T> h_img(*this->allocatorCPU.get(), this->width, this->height);
ncvAssertReturn(h_img.isMemAllocated(), false);
NCVMatrixAlloc<T> d_dst(*this->allocatorGPU.get(), this->height, this->width);
ncvAssertReturn(d_dst.isMemAllocated(), false);
NCVMatrixAlloc<T> h_dst(*this->allocatorCPU.get(), this->height, this->width);
ncvAssertReturn(h_dst.isMemAllocated(), false);
NCVMatrixAlloc<T> h_dst_d(*this->allocatorCPU.get(), this->height, this->width);
ncvAssertReturn(h_dst_d.isMemAllocated(), false);
NCV_SET_SKIP_COND(this->allocatorGPU.get()->isCounting());
NCV_SKIP_COND_BEGIN
ncvAssertReturn(this->src.fill(h_img), false);
NCV_SKIP_COND_END
ncvStat = h_img.copySolid(d_img, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStTranspose_32u_C1R((Ncv32u *)d_img.ptr(), d_img.pitch(),
(Ncv32u *)d_dst.ptr(), d_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStTranspose_64u_C1R((Ncv64u *)d_img.ptr(), d_img.pitch(),
(Ncv64u *)d_dst.ptr(), d_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else
{
ncvAssertPrintReturn(false, "Incorrect transpose test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
ncvStat = d_dst.copySolid(h_dst_d, 0);
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_BEGIN
if (sizeof(T) == sizeof(Ncv32u))
{
ncvStat = nppiStTranspose_32u_C1R_host((Ncv32u *)h_img.ptr(), h_img.pitch(),
(Ncv32u *)h_dst.ptr(), h_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else if (sizeof(T) == sizeof(Ncv64u))
{
ncvStat = nppiStTranspose_64u_C1R_host((Ncv64u *)h_img.ptr(), h_img.pitch(),
(Ncv64u *)h_dst.ptr(), h_dst.pitch(),
NcvSize32u(this->width, this->height));
}
else
{
ncvAssertPrintReturn(false, "Incorrect downsample test instance", false);
}
ncvAssertReturn(ncvStat == NPPST_SUCCESS, false);
NCV_SKIP_COND_END
//bit-to-bit check
bool bLoopVirgin = true;
NCV_SKIP_COND_BEGIN
//const Ncv64f relEPS = 0.005;
for (Ncv32u i=0; bLoopVirgin && i < this->width; i++)
{
for (Ncv32u j=0; bLoopVirgin && j < this->height; j++)
{
if (h_dst.ptr()[h_dst.stride()*i+j] != h_dst_d.ptr()[h_dst_d.stride()*i+j])
{
bLoopVirgin = false;
}
}
}
NCV_SKIP_COND_END
if (bLoopVirgin)
{
rcode = true;
}
return rcode;
}
template <class T>
bool TestTranspose<T>::deinit()
{
return true;
}
template class TestTranspose<Ncv32u>;
template class TestTranspose<Ncv64u>;
modules/cudalegacy/test/TestTranspose.h
+73
View File
@@ -0,0 +1,73 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef _testtranspose_h_
#define _testtranspose_h_
#include "NCVTest.hpp"
#include "NCVTestSourceProvider.hpp"
template <class T>
class TestTranspose : public NCVTestProvider
{
public:
TestTranspose(std::string testName, NCVTestSourceProvider<T> &src,
Ncv32u width, Ncv32u height);
virtual bool init();
virtual bool process();
virtual bool deinit();
virtual bool toString(std::ofstream &strOut);
private:
TestTranspose(const TestTranspose&);
TestTranspose& operator=(const TestTranspose&);
NCVTestSourceProvider<T> &src;
Ncv32u width;
Ncv32u height;
};
#endif // _testtranspose_h_
modules/cudalegacy/test/main_nvidia.cpp
+459
View File
@@ -0,0 +1,459 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#if defined _MSC_VER && _MSC_VER >= 1200
# pragma warning (disable : 4408 4201 4100)
#endif
static std::string path;
namespace {
template <class T_in, class T_out>
void generateIntegralTests(NCVAutoTestLister &testLister,
NCVTestSourceProvider<T_in> &src,
Ncv32u maxWidth, Ncv32u maxHeight)
{
for (Ncv32f _i=1.0; _i<maxWidth; _i*=1.2f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "LinIntImgW%dH%d", i, 2);
testLister.add(new TestIntegralImage<T_in, T_out>(testName, src, i, 2));
}
for (Ncv32f _i=1.0; _i<maxHeight; _i*=1.2f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "LinIntImgW%dH%d", 2, i);
testLister.add(new TestIntegralImage<T_in, T_out>(testName, src, 2, i));
}
testLister.add(new TestIntegralImage<T_in, T_out>("LinIntImg_VGA", src, 640, 480));
}
void generateSquaredIntegralTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<Ncv8u> &src,
Ncv32u maxWidth, Ncv32u maxHeight)
{
for (Ncv32f _i=1.0; _i<maxWidth; _i*=1.2f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "SqIntImgW%dH%d", i, 32);
testLister.add(new TestIntegralImageSquared(testName, src, i, 32));
}
for (Ncv32f _i=1.0; _i<maxHeight; _i*=1.2f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "SqIntImgW%dH%d", 32, i);
testLister.add(new TestIntegralImageSquared(testName, src, 32, i));
}
testLister.add(new TestIntegralImageSquared("SqLinIntImg_VGA", src, 640, 480));
}
void generateRectStdDevTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<Ncv8u> &src,
Ncv32u maxWidth, Ncv32u maxHeight)
{
NcvRect32u rect(1,1,18,18);
for (Ncv32f _i=32; _i<maxHeight/2 && _i < maxWidth/2; _i*=1.2f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "RectStdDevW%dH%d", i*2, i);
testLister.add(new TestRectStdDev(testName, src, i*2, i, rect, 1, true));
testLister.add(new TestRectStdDev(testName, src, i*2, i, rect, 1.5, false));
testLister.add(new TestRectStdDev(testName, src, i-1, i*2-1, rect, 1, false));
testLister.add(new TestRectStdDev(testName, src, i-1, i*2-1, rect, 2.5, true));
}
testLister.add(new TestRectStdDev("RectStdDev_VGA", src, 640, 480, rect, 1, true));
}
template <class T>
void generateResizeTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<T> &src)
{
for (Ncv32u i=2; i<10; ++i)
{
char testName[80];
sprintf(testName, "TestResize_VGA_s%d", i);
testLister.add(new TestResize<T>(testName, src, 640, 480, i, true));
testLister.add(new TestResize<T>(testName, src, 640, 480, i, false));
}
for (Ncv32u i=2; i<10; ++i)
{
char testName[80];
sprintf(testName, "TestResize_1080_s%d", i);
testLister.add(new TestResize<T>(testName, src, 1920, 1080, i, true));
testLister.add(new TestResize<T>(testName, src, 1920, 1080, i, false));
}
}
void generateNPPSTVectorTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<Ncv32u> &src, Ncv32u maxLength)
{
//compaction
for (Ncv32f _i=256.0; _i<maxLength; _i*=1.5f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "Compaction%d", i);
testLister.add(new TestCompact(testName, src, i, 0xFFFFFFFF, 30));
}
for (Ncv32u i=1; i<260; i++)
{
char testName[80];
sprintf(testName, "Compaction%d", i);
testLister.add(new TestCompact(testName, src, i, 0xC001C0DE, 70));
testLister.add(new TestCompact(testName, src, i, 0xC001C0DE, 0));
testLister.add(new TestCompact(testName, src, i, 0xC001C0DE, 100));
}
for (Ncv32u i=256*256-10; i<256*256+10; i++)
{
char testName[80];
sprintf(testName, "Compaction%d", i);
testLister.add(new TestCompact(testName, src, i, 0xFFFFFFFF, 40));
}
for (Ncv32u i=256*256*256-2; i<256*256*256+2; i++)
{
char testName[80];
sprintf(testName, "Compaction%d", i);
testLister.add(new TestCompact(testName, src, i, 0x00000000, 2));
}
}
template <class T>
void generateTransposeTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<T> &src)
{
for (int i=2; i<64; i+=4)
{
for (int j=2; j<64; j+=4)
{
char testName[80];
sprintf(testName, "TestTranspose_%dx%d", i, j);
testLister.add(new TestTranspose<T>(testName, src, i, j));
}
}
for (int i=1; i<128; i+=1)
{
for (int j=1; j<2; j+=1)
{
char testName[80];
sprintf(testName, "TestTranspose_%dx%d", i, j);
testLister.add(new TestTranspose<T>(testName, src, i, j));
}
}
testLister.add(new TestTranspose<T>("TestTranspose_VGA", src, 640, 480));
testLister.add(new TestTranspose<T>("TestTranspose_HD1080", src, 1920, 1080));
//regression tests
testLister.add(new TestTranspose<T>("TestTranspose_reg_0", src, 1072, 375));
}
template <class T>
void generateDrawRectsTests(NCVAutoTestLister &testLister,
NCVTestSourceProvider<T> &src,
NCVTestSourceProvider<Ncv32u> &src32u,
Ncv32u maxWidth, Ncv32u maxHeight)
{
for (Ncv32f _i=16.0; _i<maxWidth; _i*=1.1f)
{
Ncv32u i = (Ncv32u)_i;
Ncv32u j = maxHeight * i / maxWidth;
if (!j) continue;
char testName[80];
sprintf(testName, "DrawRectsW%dH%d", i, j);
if (sizeof(T) == sizeof(Ncv32u))
{
testLister.add(new TestDrawRects<T>(testName, src, src32u, i, j, i*j/1000+1, (T)0xFFFFFFFF));
}
else if (sizeof(T) == sizeof(Ncv8u))
{
testLister.add(new TestDrawRects<T>(testName, src, src32u, i, j, i*j/1000+1, (T)0xFF));
}
else
{
ncvAssertPrintCheck(false, "Attempted to instantiate non-existing DrawRects test suite");
}
}
//test VGA
testLister.add(new TestDrawRects<T>("DrawRects_VGA", src, src32u, 640, 480, 640*480/1000, (T)0xFF));
}
void generateVectorTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<Ncv32u> &src, Ncv32u maxLength)
{
//growth
for (Ncv32f _i=10.0; _i<maxLength; _i*=1.5f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "VectorGrow%d", i);
testLister.add(new TestHypothesesGrow(testName, src, 20, 20, 2.2f, i, i/2, i, i/4));
testLister.add(new TestHypothesesGrow(testName, src, 10, 42, 1.2f, i, i, i, 0));
}
testLister.add(new TestHypothesesGrow("VectorGrow01b", src, 10, 42, 1.2f, 10, 0, 10, 1));
testLister.add(new TestHypothesesGrow("VectorGrow11b", src, 10, 42, 1.2f, 10, 1, 10, 1));
testLister.add(new TestHypothesesGrow("VectorGrow10b", src, 10, 42, 1.2f, 10, 1, 10, 0));
testLister.add(new TestHypothesesGrow("VectorGrow00b", src, 10, 42, 1.2f, 10, 0, 10, 0));
}
void generateHypothesesFiltrationTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<Ncv32u> &src, Ncv32u maxLength)
{
for (Ncv32f _i=1.0; _i<maxLength; _i*=1.1f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "HypFilter%d", i);
testLister.add(new TestHypothesesFilter(testName, src, i, 3, 0.2f));
testLister.add(new TestHypothesesFilter(testName, src, i, 0, 0.2f));
testLister.add(new TestHypothesesFilter(testName, src, i, 1, 0.1f));
}
}
void generateHaarLoaderTests(NCVAutoTestLister &testLister)
{
testLister.add(new TestHaarCascadeLoader("haarcascade_eye.xml", path + "haarcascade_eye.xml"));
testLister.add(new TestHaarCascadeLoader("haarcascade_frontalface_alt.xml", path + "haarcascade_frontalface_alt.xml"));
testLister.add(new TestHaarCascadeLoader("haarcascade_frontalface_alt2.xml", path + "haarcascade_frontalface_alt2.xml"));
testLister.add(new TestHaarCascadeLoader("haarcascade_frontalface_alt_tree.xml", path + "haarcascade_frontalface_alt_tree.xml"));
testLister.add(new TestHaarCascadeLoader("haarcascade_eye_tree_eyeglasses.xml", path + "haarcascade_eye_tree_eyeglasses.xml"));
}
void generateHaarApplicationTests(NCVAutoTestLister &testLister, NCVTestSourceProvider<Ncv8u> &src,
Ncv32u maxWidth, Ncv32u maxHeight)
{
CV_UNUSED(maxHeight);
for (Ncv32u i=100; i<512; i+=41)
{
for (Ncv32u j=100; j<128; j+=25)
{
char testName[80];
sprintf(testName, "HaarAppl%d_%d", i, j);
testLister.add(new TestHaarCascadeApplication(testName, src, path + "haarcascade_frontalface_alt.xml", j, i));
}
}
for (Ncv32f _i=20.0; _i<maxWidth; _i*=1.5f)
{
Ncv32u i = (Ncv32u)_i;
char testName[80];
sprintf(testName, "HaarAppl%d", i);
testLister.add(new TestHaarCascadeApplication(testName, src, path + "haarcascade_frontalface_alt.xml", i, i));
}
}
static void devNullOutput(const cv::String& msg)
{
CV_UNUSED(msg);
}
}
bool nvidia_NPPST_Integral_Image(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path.c_str();
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerII("NPPST Integral Image", outputLevel);
NCVTestSourceProvider<Ncv8u> testSrcRandom_8u(2010, 0, 255, 2048, 2048);
NCVTestSourceProvider<Ncv32f> testSrcRandom_32f(2010, -1.0f, 1.0f, 2048, 2048);
generateIntegralTests<Ncv8u, Ncv32u>(testListerII, testSrcRandom_8u, 2048, 2048);
generateIntegralTests<Ncv32f, Ncv32f>(testListerII, testSrcRandom_32f, 2048, 2048);
return testListerII.invoke();
}
bool nvidia_NPPST_Squared_Integral_Image(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerSII("NPPST Squared Integral Image", outputLevel);
NCVTestSourceProvider<Ncv8u> testSrcRandom_8u(2010, 0, 255, 2048, 2048);
generateSquaredIntegralTests(testListerSII, testSrcRandom_8u, 2048, 2048);
return testListerSII.invoke();
}
bool nvidia_NPPST_RectStdDev(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerRStdDev("NPPST RectStdDev", outputLevel);
NCVTestSourceProvider<Ncv8u> testSrcRandom_8u(2010, 0, 255, 2048, 2048);
generateRectStdDevTests(testListerRStdDev, testSrcRandom_8u, 2048, 2048);
return testListerRStdDev.invoke();
}
bool nvidia_NPPST_Resize(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerResize("NPPST Resize", outputLevel);
NCVTestSourceProvider<Ncv32u> testSrcRandom_32u(2010, 0, 0xFFFFFFFF, 2048, 2048);
NCVTestSourceProvider<Ncv64u> testSrcRandom_64u(2010, 0, (Ncv64u) -1, 2048, 2048);
generateResizeTests(testListerResize, testSrcRandom_32u);
generateResizeTests(testListerResize, testSrcRandom_64u);
return testListerResize.invoke();
}
bool nvidia_NPPST_Vector_Operations(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerNPPSTVectorOperations("NPPST Vector Operations", outputLevel);
NCVTestSourceProvider<Ncv32u> testSrcRandom_32u(2010, 0, 0xFFFFFFFF, 2048, 2048);
generateNPPSTVectorTests(testListerNPPSTVectorOperations, testSrcRandom_32u, 2048*2048);
return testListerNPPSTVectorOperations.invoke();
}
bool nvidia_NPPST_Transpose(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerTranspose("NPPST Transpose", outputLevel);
NCVTestSourceProvider<Ncv32u> testSrcRandom_32u(2010, 0, 0xFFFFFFFF, 2048, 2048);
NCVTestSourceProvider<Ncv64u> testSrcRandom_64u(2010, 0, (Ncv64u) -1, 2048, 2048);
generateTransposeTests(testListerTranspose, testSrcRandom_32u);
generateTransposeTests(testListerTranspose, testSrcRandom_64u);
return testListerTranspose.invoke();
}
bool nvidia_NCV_Vector_Operations(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerVectorOperations("Vector Operations", outputLevel);
NCVTestSourceProvider<Ncv32u> testSrcRandom_32u(2010, 0, 0xFFFFFFFF, 2048, 2048);
generateVectorTests(testListerVectorOperations, testSrcRandom_32u, 2048*2048);
return testListerVectorOperations.invoke();
}
bool nvidia_NCV_Haar_Cascade_Loader(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerHaarLoader("Haar Cascade Loader", outputLevel);
generateHaarLoaderTests(testListerHaarLoader);
return testListerHaarLoader.invoke();
}
bool nvidia_NCV_Haar_Cascade_Application(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerHaarAppl("Haar Cascade Application", outputLevel);
NCVTestSourceProvider<Ncv8u> testSrcFacesVGA_8u(path + "group_1_640x480_VGA.pgm");
generateHaarApplicationTests(testListerHaarAppl, testSrcFacesVGA_8u, 640, 480);
return testListerHaarAppl.invoke();
}
bool nvidia_NCV_Hypotheses_Filtration(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerHypFiltration("Hypotheses Filtration", outputLevel);
NCVTestSourceProvider<Ncv32u> testSrcRandom_32u(2010, 0, 0xFFFFFFFF, 2048, 2048);
generateHypothesesFiltrationTests(testListerHypFiltration, testSrcRandom_32u, 512);
return testListerHypFiltration.invoke();
}
bool nvidia_NCV_Visualization(const std::string& test_data_path, OutputLevel outputLevel)
{
path = test_data_path;
ncvSetDebugOutputHandler(devNullOutput);
NCVAutoTestLister testListerVisualize("Visualization", outputLevel);
NCVTestSourceProvider<Ncv8u> testSrcRandom_8u(2010, 0, 255, 2048, 2048);
NCVTestSourceProvider<Ncv32u> testSrcRandom_32u(2010, 0, RAND_MAX, 2048, 2048);
generateDrawRectsTests(testListerVisualize, testSrcRandom_8u, testSrcRandom_32u, 2048, 2048);
generateDrawRectsTests(testListerVisualize, testSrcRandom_32u, testSrcRandom_32u, 2048, 2048);
return testListerVisualize.invoke();
}
modules/cudalegacy/test/main_test_nvidia.h
+67
View File
@@ -0,0 +1,67 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __main_test_nvidia_h__
#define __main_test_nvidia_h__
enum OutputLevel
{
OutputLevelNone,
OutputLevelCompact,
OutputLevelFull
};
extern OutputLevel nvidiaTestOutputLevel;
bool nvidia_NPPST_Integral_Image(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NPPST_Squared_Integral_Image(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NPPST_RectStdDev(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NPPST_Resize(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NPPST_Vector_Operations(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NPPST_Transpose(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NCV_Vector_Operations(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NCV_Haar_Cascade_Loader(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NCV_Haar_Cascade_Application(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NCV_Hypotheses_Filtration(const std::string& test_data_path, OutputLevel outputLevel);
bool nvidia_NCV_Visualization(const std::string& test_data_path, OutputLevel outputLevel);
#endif
modules/cudalegacy/test/test_calib3d.cpp
+193
View File
@@ -0,0 +1,193 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#if defined HAVE_CUDA && defined HAVE_OPENCV_3D
#include "opencv2/geometry.hpp"
namespace opencv_test { namespace {
///////////////////////////////////////////////////////////////////////////////////////////////////////
// transformPoints
struct TransformPoints : testing::TestWithParam<cv::cuda::DeviceInfo>
{
cv::cuda::DeviceInfo devInfo;
virtual void SetUp()
{
devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(TransformPoints, Accuracy)
{
cv::Mat src = randomMat(cv::Size(1000, 1), CV_32FC3, 0, 10);
cv::Mat rvec = randomMat(cv::Size(3, 1), CV_32F, 0, 1);
cv::Mat tvec = randomMat(cv::Size(3, 1), CV_32F, 0, 1);
cv::cuda::GpuMat dst;
cv::cuda::transformPoints(loadMat(src), rvec, tvec, dst);
ASSERT_EQ(src.size(), dst.size());
ASSERT_EQ(src.type(), dst.type());
cv::Mat h_dst(dst);
cv::Mat rot;
cv::Rodrigues(rvec, rot);
for (int i = 0; i < h_dst.cols; ++i)
{
cv::Point3f res = h_dst.at<cv::Point3f>(0, i);
cv::Point3f p = src.at<cv::Point3f>(0, i);
cv::Point3f res_gold(
rot.at<float>(0, 0) * p.x + rot.at<float>(0, 1) * p.y + rot.at<float>(0, 2) * p.z + tvec.at<float>(0, 0),
rot.at<float>(1, 0) * p.x + rot.at<float>(1, 1) * p.y + rot.at<float>(1, 2) * p.z + tvec.at<float>(0, 1),
rot.at<float>(2, 0) * p.x + rot.at<float>(2, 1) * p.y + rot.at<float>(2, 2) * p.z + tvec.at<float>(0, 2));
ASSERT_POINT3_NEAR(res_gold, res, 1e-5);
}
}
INSTANTIATE_TEST_CASE_P(CUDA_Calib3D, TransformPoints, ALL_DEVICES);
///////////////////////////////////////////////////////////////////////////////////////////////////////
// ProjectPoints
struct ProjectPoints : testing::TestWithParam<cv::cuda::DeviceInfo>
{
cv::cuda::DeviceInfo devInfo;
virtual void SetUp()
{
devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(ProjectPoints, Accuracy)
{
cv::Mat src = randomMat(cv::Size(1000, 1), CV_32FC3, 0, 10);
cv::Mat rvec = randomMat(cv::Size(3, 1), CV_32F, 0, 1);
cv::Mat tvec = randomMat(cv::Size(3, 1), CV_32F, 0, 1);
cv::Mat camera_mat = randomMat(cv::Size(3, 3), CV_32F, 0.5, 1);
camera_mat.at<float>(0, 1) = 0.f;
camera_mat.at<float>(1, 0) = 0.f;
camera_mat.at<float>(2, 0) = 0.f;
camera_mat.at<float>(2, 1) = 0.f;
cv::cuda::GpuMat dst;
cv::cuda::projectPoints(loadMat(src), rvec, tvec, camera_mat, cv::Mat(), dst);
ASSERT_EQ(1, dst.rows);
ASSERT_EQ(MatType(CV_32FC2), MatType(dst.type()));
std::vector<cv::Point2f> dst_gold;
cv::projectPoints(src, rvec, tvec, camera_mat, cv::Mat(1, 8, CV_32F, cv::Scalar::all(0)), dst_gold);
ASSERT_EQ(dst_gold.size(), static_cast<size_t>(dst.cols));
cv::Mat h_dst(dst);
for (size_t i = 0; i < dst_gold.size(); ++i)
{
cv::Point2f res = h_dst.at<cv::Point2f>(0, (int)i);
cv::Point2f res_gold = dst_gold[i];
ASSERT_LE(cv::norm(res_gold - res) / cv::norm(res_gold), 1e-3f);
}
}
INSTANTIATE_TEST_CASE_P(CUDA_Calib3D, ProjectPoints, ALL_DEVICES);
///////////////////////////////////////////////////////////////////////////////////////////////////////
// SolvePnPRansac
struct SolvePnPRansac : testing::TestWithParam<cv::cuda::DeviceInfo>
{
cv::cuda::DeviceInfo devInfo;
virtual void SetUp()
{
devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
}
};
CUDA_TEST_P(SolvePnPRansac, Accuracy)
{
cv::Mat object = randomMat(cv::Size(5000, 1), CV_32FC3, -2000, 2000);
cv::Mat camera_mat = randomMat(cv::Size(3, 3), CV_32F, 0.5, 1);
camera_mat.at<float>(0, 1) = 0.f;
camera_mat.at<float>(1, 0) = 0.f;
camera_mat.at<float>(2, 0) = 0.f;
camera_mat.at<float>(2, 1) = 0.f;
std::vector<cv::Point2f> image_vec;
cv::Mat rvec_gold;
cv::Mat tvec_gold;
rvec_gold = randomMat(cv::Size(3, 1), CV_32F, 0, 1);
tvec_gold = randomMat(cv::Size(3, 1), CV_32F, 0, 1000);
cv::projectPoints(object, rvec_gold, tvec_gold, camera_mat, cv::Mat(1, 8, CV_32F, cv::Scalar::all(0)), image_vec);
cv::Mat rvec, tvec;
std::vector<int> inliers;
cv::cuda::solvePnPRansac(object, cv::Mat(1, (int)image_vec.size(), CV_32FC2, &image_vec[0]),
camera_mat, cv::Mat(1, 8, CV_32F, cv::Scalar::all(0)),
rvec, tvec, false, 200, 2.f, 100, &inliers);
ASSERT_LE(cv::norm(rvec - rvec_gold), 1e-3);
ASSERT_LE(cv::norm(tvec, tvec_gold, NORM_L2 | NORM_RELATIVE), 1e-3);
}
INSTANTIATE_TEST_CASE_P(CUDA_Calib3D, SolvePnPRansac, ALL_DEVICES);
}} // namespace
#endif // HAVE_CUDA
modules/cudalegacy/test/test_labeling.cpp
+200
View File
@@ -0,0 +1,200 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
namespace opencv_test { namespace {
namespace
{
struct GreedyLabeling
{
struct dot
{
int x;
int y;
static dot make(int i, int j)
{
dot d; d.x = i; d.y = j;
return d;
}
};
struct InInterval
{
InInterval(const int& _lo, const int& _hi) : lo(-_lo), hi(_hi) {}
const int lo, hi;
bool operator() (const unsigned char a, const unsigned char b) const
{
int d = a - b;
return lo <= d && d <= hi;
}
};
GreedyLabeling(cv::Mat img)
: image(img), _labels(image.size(), CV_32SC1, cv::Scalar::all(-1)) {}
void operator() (cv::Mat labels) const
{
InInterval inInt(0, 2);
dot* stack = new dot[image.cols * image.rows];
int cc = -1;
int* dist_labels = (int*)labels.data;
int pitch = (int) labels.step1();
unsigned char* source = (unsigned char*)image.data;
int width = image.cols;
int height = image.rows;
int step1 = (int)image.step1();
for (int j = 0; j < image.rows; ++j)
for (int i = 0; i < image.cols; ++i)
{
if (dist_labels[j * pitch + i] != -1) continue;
dot* top = stack;
dot p = dot::make(i, j);
cc++;
dist_labels[j * pitch + i] = cc;
while (top >= stack)
{
int* dl = &dist_labels[p.y * pitch + p.x];
unsigned char* sp = &source[p.y * step1 + p.x];
dl[0] = cc;
//right
if( p.x < (width - 1) && dl[ +1] == -1 && inInt(sp[0], sp[+1]))
*top++ = dot::make(p.x + 1, p.y);
//left
if( p.x > 0 && dl[-1] == -1 && inInt(sp[0], sp[-1]))
*top++ = dot::make(p.x - 1, p.y);
//bottom
if( p.y < (height - 1) && dl[+pitch] == -1 && inInt(sp[0], sp[+step1]))
*top++ = dot::make(p.x, p.y + 1);
//top
if( p.y > 0 && dl[-pitch] == -1 && inInt(sp[0], sp[-step1]))
*top++ = dot::make(p.x, p.y - 1);
p = *--top;
}
}
delete[] stack;
}
void checkCorrectness(cv::Mat gpu)
{
cv::Mat diff = gpu - _labels;
int outliers = 0;
for (int j = 0; j < image.rows; ++j)
for (int i = 0; i < image.cols - 1; ++i)
{
if ( (_labels.at<int>(j,i) == gpu.at<int>(j,i + 1)) && (diff.at<int>(j, i) != diff.at<int>(j,i + 1)))
{
outliers++;
}
}
ASSERT_TRUE(outliers < gpu.cols + gpu.rows);
}
cv::Mat image;
cv::Mat _labels;
};
}
struct Labeling : testing::TestWithParam<cv::cuda::DeviceInfo>
{
cv::cuda::DeviceInfo devInfo;
virtual void SetUp()
{
devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
}
cv::Mat loat_image()
{
return cv::imread(std::string( cvtest::TS::ptr()->get_data_path() ) + "labeling/label.png");
}
};
CUDA_TEST_P(Labeling, DISABLED_ConnectedComponents)
{
cv::Mat image;
cvtColor(loat_image(), image, cv::COLOR_BGR2GRAY);
cv::threshold(image, image, 150, 255, cv::THRESH_BINARY);
ASSERT_TRUE(image.type() == CV_8UC1);
GreedyLabeling host(image);
host(host._labels);
cv::cuda::GpuMat mask;
mask.create(image.rows, image.cols, CV_8UC1);
cv::cuda::GpuMat components;
components.create(image.rows, image.cols, CV_32SC1);
cv::cuda::connectivityMask(cv::cuda::GpuMat(image), mask, cv::Scalar::all(0), cv::Scalar::all(2));
cv::cuda::labelComponents(mask, components);
host.checkCorrectness(cv::Mat(components));
}
INSTANTIATE_TEST_CASE_P(CUDA_ConnectedComponents, Labeling, ALL_DEVICES);
}} // namespace
#endif // HAVE_CUDA
modules/cudalegacy/test/test_main.cpp
+130
View File
@@ -0,0 +1,130 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
using namespace std;
using namespace cv;
using namespace cv::cuda;
using namespace opencv_test;
using namespace testing;
int main(int argc, char** argv)
{
try
{
const std::string keys =
"{ h help ? | | Print help}"
"{ i info | | Print information about system and exit }"
"{ device | -1 | Device on which tests will be executed (-1 means all devices) }"
"{ nvtest_output_level | none | NVidia test verbosity level (none, compact, full) }"
;
CommandLineParser cmd(argc, (const char**)argv, keys);
if (cmd.has("help"))
{
cmd.printMessage();
return 0;
}
printCudaInfo();
if (cmd.has("info"))
{
return 0;
}
int device = cmd.get<int>("device");
if (device < 0)
{
DeviceManager::instance().loadAll();
cout << "Run tests on all supported devices \n" << endl;
}
else
{
DeviceManager::instance().load(device);
DeviceInfo info(device);
cout << "Run tests on device " << device << " [" << info.name() << "] \n" << endl;
}
string outputLevel = cmd.get<string>("nvtest_output_level");
if (outputLevel == "none")
nvidiaTestOutputLevel = OutputLevelNone;
else if (outputLevel == "compact")
nvidiaTestOutputLevel = OutputLevelCompact;
else if (outputLevel == "full")
nvidiaTestOutputLevel = OutputLevelFull;
TS::ptr()->init("gpu");
InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}
catch (const exception& e)
{
cerr << e.what() << endl;
return -1;
}
catch (...)
{
cerr << "Unknown error" << endl;
return -1;
}
return 0;
}
#else // HAVE_CUDA
int main()
{
printf("OpenCV was built without CUDA support\n");
return 0;
}
#endif // HAVE_CUDA
modules/cudalegacy/test/test_nvidia.cpp
+152
View File
@@ -0,0 +1,152 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#ifdef HAVE_CUDA
OutputLevel nvidiaTestOutputLevel = OutputLevelNone;
namespace opencv_test { namespace {
struct NVidiaTest : TestWithParam<cv::cuda::DeviceInfo>
{
cv::cuda::DeviceInfo devInfo;
std::string _path;
virtual void SetUp()
{
devInfo = GetParam();
cv::cuda::setDevice(devInfo.deviceID());
_path = TS::ptr()->get_data_path().c_str();
_path = _path + "haarcascade/";
}
};
struct NPPST : NVidiaTest {};
struct NCV : NVidiaTest {};
CUDA_TEST_P(NPPST, Integral)
{
bool res = nvidia_NPPST_Integral_Image(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NPPST, SquaredIntegral)
{
bool res = nvidia_NPPST_Squared_Integral_Image(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NPPST, RectStdDev)
{
bool res = nvidia_NPPST_RectStdDev(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NPPST, Resize)
{
bool res = nvidia_NPPST_Resize(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NPPST, VectorOperations)
{
bool res = nvidia_NPPST_Vector_Operations(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NPPST, Transpose)
{
bool res = nvidia_NPPST_Transpose(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NCV, VectorOperations)
{
bool res = nvidia_NCV_Vector_Operations(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NCV, HaarCascadeLoader)
{
bool res = nvidia_NCV_Haar_Cascade_Loader(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NCV, HaarCascadeApplication)
{
bool res = nvidia_NCV_Haar_Cascade_Application(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NCV, HypothesesFiltration)
{
bool res = nvidia_NCV_Hypotheses_Filtration(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
CUDA_TEST_P(NCV, Visualization)
{
bool res = nvidia_NCV_Visualization(_path, nvidiaTestOutputLevel);
ASSERT_TRUE(res);
}
INSTANTIATE_TEST_CASE_P(CUDA_Legacy, NPPST, ALL_DEVICES);
INSTANTIATE_TEST_CASE_P(CUDA_Legacy, NCV, ALL_DEVICES);
}} // namespace
#endif // HAVE_CUDA
modules/cudalegacy/test/test_precomp.hpp
+90
View File
@@ -0,0 +1,90 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_TEST_PRECOMP_HPP__
#define __OPENCV_TEST_PRECOMP_HPP__
#if defined(__GNUC__) && !defined(__APPLE__) && !defined(__arm__) && !defined(__aarch64__) && !defined(__powerpc64__)
#include <fpu_control.h>
#endif
#include <cfloat>
#include <cstdio>
#include <cmath>
#include <vector>
#include <string>
#include <map>
#include <memory>
#include <algorithm>
#include <fstream>
#include "opencv2/ts.hpp"
#include "opencv2/ts/cuda_test.hpp"
#include "opencv2/core/cuda.hpp"
#include "opencv2/cudalegacy.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/core/private.cuda.hpp"
#include "opencv2/opencv_modules.hpp"
#include "cvconfig.h"
#include "NCVTest.hpp"
#include "NCVAutoTestLister.hpp"
#include "NCVTestSourceProvider.hpp"
#include "TestIntegralImage.h"
#include "TestIntegralImageSquared.h"
#include "TestRectStdDev.h"
#include "TestResize.h"
#include "TestCompact.h"
#include "TestTranspose.h"
#include "TestDrawRects.h"
#include "TestHypothesesGrow.h"
#include "TestHypothesesFilter.h"
#include "TestHaarCascadeLoader.h"
#include "TestHaarCascadeApplication.h"
#include "main_test_nvidia.h"
#endif