Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge branch 2.4

Browse Source
pull/534/merge
Andrey Kamaev 12 years ago
parent 0b8a6da817 620c699456
commit dddd285d3b
29 changed files with 226 additions and 2418 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 4
      doc/tutorials/core/adding_images/adding_images.rst
  2. 176
      modules/ocl/src/arithm.cpp
  3. 12
      modules/ocl/src/binarycaching.hpp
  4. 8
      modules/ocl/src/blend.cpp
  5. 134
      modules/ocl/src/brute_force_matcher.cpp
  6. 18
      modules/ocl/src/build_warps.cpp
  7. 21
      modules/ocl/src/canny.cpp
  8. 14
      modules/ocl/src/color.cpp
  9. 11
      modules/ocl/src/columnsum.cpp
  10. 8
      modules/ocl/src/error.cpp
  11. 7
      modules/ocl/src/fft.cpp
  12. 117
      modules/ocl/src/filtering.cpp
  13. 19
      modules/ocl/src/gemm.cpp
  14. 1256
      modules/ocl/src/haar.cpp
  15. 65
      modules/ocl/src/hog.cpp
  16. 57
      modules/ocl/src/imgproc.cpp
  17. 77
      modules/ocl/src/initialization.cpp
  18. 12
      modules/ocl/src/interpolate_frames.cpp
  19. 409
      modules/ocl/src/kernels/haarobjectdetect.cl
  20. 7
      modules/ocl/src/match_template.cpp
  21. 55
      modules/ocl/src/matrix_operations.cpp
  22. 5
      modules/ocl/src/mcwutil.hpp
  23. 18
      modules/ocl/src/mssegmentation.cpp
  24. 9
      modules/ocl/src/pyrlk.cpp
  25. 10
      modules/ocl/src/pyrup.cpp
  26. 29
      modules/ocl/src/split_merge.cpp
  27. 55
      modules/ocl/src/surf.cpp
  28. 4
      modules/ocl/test/test_imgproc.cpp
  29. 21
      modules/ocl/test/test_match_template.cpp

4
doc/tutorials/core/adding_images/adding_images.rst
Unescape View File

@ -53,8 +53,8 @@ As usual, after the not-so-lengthy explanation, let's go to the code:
std::cout<<"* Enter alpha [0-1]: ";
std::cin>>input;
/// We use the alpha provided by the user iff it is between 0 and 1
if( alpha >= 0 && alpha <= 1 )
/// We use the alpha provided by the user if it is between 0 and 1
if( input >= 0.0 && input <= 1.0 )
{ alpha = input; }
/// Read image ( same size, same type )

176
modules/ocl/src/arithm.cpp
Unescape View File

@ -57,180 +57,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
/* arithmetics */
void cv::ocl::add(const oclMat &, const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::add(const oclMat &, const oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::add(const oclMat &, const Scalar &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::subtract(const oclMat &, const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::subtract(const oclMat &, const oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::subtract(const oclMat &, const Scalar &, oclMat &, const oclMat & )
{
throw_nogpu();
}
void cv::ocl::subtract(const Scalar &, const oclMat &, oclMat &, const oclMat & )
{
throw_nogpu();
}
void cv::ocl::multiply(const oclMat &, const oclMat &, oclMat &, double)
{
throw_nogpu();
}
void cv::ocl::divide(const oclMat &, const oclMat &, oclMat &, double)
{
throw_nogpu();
}
void cv::ocl::divide(double, const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::absdiff(const oclMat &, const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::absdiff(const oclMat &, const Scalar &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::compare(const oclMat &, const oclMat &, oclMat & , int)
{
throw_nogpu();
}
void cv::ocl::meanStdDev(const oclMat &, Scalar &, Scalar &)
{
throw_nogpu();
}
double cv::ocl::norm(const oclMat &, int)
{
throw_nogpu();
return 0.0;
}
double cv::ocl::norm(const oclMat &, const oclMat &, int)
{
throw_nogpu();
return 0.0;
}
void cv::ocl::flip(const oclMat &, oclMat &, int)
{
throw_nogpu();
}
Scalar cv::ocl::sum(const oclMat &)
{
throw_nogpu();
return Scalar();
}
void cv::ocl::minMax(const oclMat &, double *, double *, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::minMaxLoc(const oclMat &, double *, double *, Point *, Point *, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::LUT(const oclMat &, const Mat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::exp(const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::log(const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::magnitude(const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::phase(const oclMat &, const oclMat &, oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::cartToPolar(const oclMat &, const oclMat &, oclMat &, oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::polarToCart(const oclMat &, const oclMat &, oclMat &, oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::transpose(const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::pow(const oclMat &, double, oclMat &)
{
throw_nogpu();
}
void cv::ocl::addWeighted(const oclMat &src1, double alpha, const oclMat &src2, double beta, double gama, oclMat &dst)
{
throw_nogpu();
}
void cv::ocl::magnitudeSqr(const oclMat &src1, const oclMat &src2, oclMat &dst)
{
throw_nogpu();
}
/* bit wise operations */
void cv::ocl::bitwise_not(const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::bitwise_or(const oclMat &, const oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::bitwise_and(const oclMat &, const oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::bitwise_and(const oclMat &, const Scalar &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::bitwise_xor(const oclMat &, const oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
cv::ocl::oclMat cv::ocl::operator ~ (const oclMat &)
{
throw_nogpu();
return oclMat();
}
cv::ocl::oclMat cv::ocl::operator | (const oclMat &, const oclMat &)
{
throw_nogpu();
return oclMat();
}
cv::ocl::oclMat cv::ocl::operator & (const oclMat &, const oclMat &)
{
throw_nogpu();
return oclMat();
}
cv::ocl::oclMat cv::ocl::operator ^ (const oclMat &, const oclMat &)
{
throw_nogpu();
return oclMat();
}
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -2477,5 +2303,3 @@ void cv::ocl::pow(const oclMat &x, double p, oclMat &y)
arithmetic_pow_run(x, p, y, kernelName, &arithm_pow);
}
#endif /* !defined (HAVE_OPENCL) */

12
modules/ocl/src/binarycaching.hpp
Unescape View File

@ -50,17 +50,6 @@ using namespace std;
using std::cout;
using std::endl;
#if !defined (HAVE_OPENCL)
namespace cv
{
namespace ocl
{
//nothing
}//namespace ocl
}//namespace cv
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -99,4 +88,3 @@ namespace cv
}//namespace ocl
}//namespace cv
#endif

8
modules/ocl/src/blend.cpp
Unescape View File

@ -50,13 +50,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
void cv::ocl::blendLinear(const oclMat &img1, const oclMat &img2, const oclMat &weights1, const oclMat &weights2,
oclMat &result)
{
throw_nogpu();
}
#else
namespace cv
{
namespace ocl
@ -98,4 +91,3 @@ void cv::ocl::blendLinear(const oclMat &img1, const oclMat &img2, const oclMat &
openCLExecuteKernel(ctx, &blend_linear, kernelName, globalSize, localSize, args, channels, depth);
}
}
#endif

134
modules/ocl/src/brute_force_matcher.cpp
Unescape View File

@ -51,136 +51,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
cv::ocl::BruteForceMatcher_OCL_base::BruteForceMatcher_OCL_base(DistType)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::add(const vector<oclMat> &)
{
throw_nogpu();
}
const vector<oclMat> &cv::ocl::BruteForceMatcher_OCL_base::getTrainDescriptors() const
{
throw_nogpu();
return trainDescCollection;
}
void cv::ocl::BruteForceMatcher_OCL_base::clear()
{
throw_nogpu();
}
bool cv::ocl::BruteForceMatcher_OCL_base::empty() const
{
throw_nogpu();
return true;
}
bool cv::ocl::BruteForceMatcher_OCL_base::isMaskSupported() const
{
throw_nogpu();
return true;
}
void cv::ocl::BruteForceMatcher_OCL_base::matchSingle(const oclMat &, const oclMat &, oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::matchDownload(const oclMat &, const oclMat &, vector<DMatch> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::matchConvert(const Mat &, const Mat &, vector<DMatch> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::match(const oclMat &, const oclMat &, vector<DMatch> &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::makeGpuCollection(oclMat &, oclMat &, const vector<oclMat> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::matchCollection(const oclMat &, const oclMat &, oclMat &, oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::matchDownload(const oclMat &, const oclMat &, const oclMat &, vector<DMatch> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::matchConvert(const Mat &, const Mat &, const Mat &, vector<DMatch> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::match(const oclMat &, vector<DMatch> &, const vector<oclMat> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatchSingle(const oclMat &, const oclMat &, oclMat &, oclMat &, oclMat &, int, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatchDownload(const oclMat &, const oclMat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatchConvert(const Mat &, const Mat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatch(const oclMat &, const oclMat &, vector< vector<DMatch> > &, int, const oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatch2Collection(const oclMat &, const oclMat &, oclMat &, oclMat &, oclMat &, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatch2Download(const oclMat &, const oclMat &, const oclMat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatch2Convert(const Mat &, const Mat &, const Mat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::knnMatch(const oclMat &, vector< vector<DMatch> > &, int, const vector<oclMat> &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchSingle(const oclMat &, const oclMat &, oclMat &, oclMat &, oclMat &, float, const oclMat &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchDownload(const oclMat &, const oclMat &, const oclMat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchConvert(const Mat &, const Mat &, const Mat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatch(const oclMat &, const oclMat &, vector< vector<DMatch> > &, float, const oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchCollection(const oclMat &, oclMat &, oclMat &, oclMat &, oclMat &, float, const vector<oclMat> &)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchDownload(const oclMat &, const oclMat &, const oclMat &, const oclMat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatchConvert(const Mat &, const Mat &, const Mat &, const Mat &, vector< vector<DMatch> > &, bool)
{
throw_nogpu();
}
void cv::ocl::BruteForceMatcher_OCL_base::radiusMatch(const oclMat &, vector< vector<DMatch> > &, float, const vector<oclMat> &, bool)
{
throw_nogpu();
}
#else /* !defined (HAVE_OPENCL) */
using namespace std;
namespace cv
{
@ -1827,7 +1697,3 @@ void cv::ocl::BruteForceMatcher_OCL_base::radiusMatch(const oclMat &query, vecto
radiusMatchCollection(query, trainIdx, imgIdx, distance, nMatches, maxDistance, masks);
radiusMatchDownload(trainIdx, imgIdx, distance, nMatches, matches, compactResult);
}
#endif

18
modules/ocl/src/build_warps.cpp
Unescape View File

@ -49,21 +49,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
void cv::ocl::buildWarpPlaneMaps(Size, Rect, const Mat &, const Mat &, const Mat &, float, oclMat &, oclMat &, Stream &)
{
throw_nogpu();
}
void cv::ocl::buildWarpCylindricalMaps(Size, Rect, const Mat &, const Mat &, float, oclMat &, oclMat &, Stream &)
{
throw_nogpu();
}
void cv::ocl::buildWarpSphericalMaps(Size, Rect, const Mat &, const Mat &, float, oclMat &, oclMat &, Stream &)
{
throw_nogpu();
}
#else
namespace cv
{
namespace ocl
@ -275,6 +260,3 @@ void cv::ocl::buildWarpPerspectiveMaps(const Mat &M, bool inverse, Size dsize, o
size_t localThreads[3] = {32, 8, 1};
openCLExecuteKernel(clCxt, &build_warps, kernelName, globalThreads, localThreads, args, -1, -1);
}
#endif // HAVE_OPENCL

21
modules/ocl/src/canny.cpp
Unescape View File

@ -51,25 +51,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
void cv::ocl::Canny(const oclMat &image, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false)
{
throw_nogpu();
}
void cv::ocl::Canny(const oclMat &image, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, int apperture_size = 3, bool L2gradient = false)
{
throw_nogpu();
}
void cv::ocl::Canny(const oclMat &dx, const oclMat &dy, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false)
{
throw_nogpu();
}
void cv::ocl::Canny(const oclMat &dx, const oclMat &dy, CannyBuf &buf, oclMat &edges, double low_thresh, double high_thresh, bool L2gradient = false)
{
throw_nogpu();
}
#else
namespace cv
{
namespace ocl
@ -426,5 +407,3 @@ void canny::getEdges_gpu(oclMat &map, oclMat &dst, int rows, int cols)
openCLExecuteKernel2(clCxt, &imgproc_canny, kernelName, globalThreads, localThreads, args, -1, -1);
}
#endif // HAVE_OPENCL

14
modules/ocl/src/color.cpp
Unescape View File

@ -49,18 +49,6 @@
using namespace cv;
using namespace cv::ocl;
#if !defined (HAVE_OPENCL)
void cv::ocl::cvtColor(const oclMat &, oclMat &, int, int)
{
throw_nogpu();
}
void cv::ocl::cvtColor(const oclMat &, oclMat &, int, int, const Stream &)
{
throw_nogpu();
}
#else /* !defined (HAVE_OPENCL) */
#ifndef CV_DESCALE
#define CV_DESCALE(x, n) (((x) + (1 << ((n)-1))) >> (n))
#endif
@ -289,5 +277,3 @@ void cv::ocl::cvtColor(const oclMat &src, oclMat &dst, int code, int dcn)
{
cvtColor_caller(src, dst, code, dcn);
}
#endif /* !defined (HAVE_OPENCL) */

11
modules/ocl/src/columnsum.cpp
Unescape View File

@ -50,16 +50,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined(HAVE_OPENCL)
void cv::ocl::columnSum(const oclMat &src, oclMat &dst)
{
throw_nogpu();
}
#else /*!HAVE_OPENCL */
namespace cv
{
namespace ocl
@ -93,4 +83,3 @@ void cv::ocl::columnSum(const oclMat &src, oclMat &dst)
openCLExecuteKernel(clCxt, &imgproc_columnsum, kernelName, globalThreads, localThreads, args, src.channels(), src.depth());
}
#endif

8
modules/ocl/src/error.cpp
Unescape View File

@ -46,12 +46,6 @@
using namespace cv;
using namespace cv::ocl;
#if !defined (HAVE_OPENCL)
// do nothing
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -194,5 +188,3 @@ namespace cv
}
}
}
#endif

7
modules/ocl/src/fft.cpp
Unescape View File

@ -49,12 +49,7 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined HAVE_OPENCL
void cv::ocl::dft(const oclMat &, oclMat &, Size , int )
{
throw_nogpu();
}
#elif !defined HAVE_CLAMDFFT
#if !defined HAVE_CLAMDFFT
void cv::ocl::dft(const oclMat&, oclMat&, Size, int)
{
CV_Error(CV_StsNotImplemented, "OpenCL DFT is not implemented");

117
modules/ocl/src/filtering.cpp
Unescape View File

@ -54,121 +54,6 @@ using namespace std;
using namespace cv;
using namespace cv::ocl;
#if !defined (HAVE_OPENCL)
Ptr<BaseFilter_GPU> cv::ocl::getBoxFilter_GPU(int, int, const Size &, Point, int)
{
throw_nogpu();
return Ptr<BaseFilter_GPU>(0);
}
Ptr<FilterEngine_GPU> cv::ocl::createBoxFilter_GPU(int, int, const Size &, const Point &, int)
{
throw_nogpu();
return Ptr<FilterEngine_GPU>(0);
}
Ptr<FilterEngine_GPU> cv::ocl::createFilter2D_GPU(const Ptr<BaseFilter_GPU>)
{
throw_nogpu();
return Ptr<FilterEngine_GPU>(0);
}
Ptr<FilterEngine_GPU> cv::ocl::createSeparableLinearFilter_GPU(int, int, const Mat &, const Mat &, const Point &)
{
throw_nogpu();
return Ptr<FilterEngine_GPU>(0);
}
Ptr<BaseRowFilter_GPU> cv::ocl::getLinearRowFilter_GPU(int, int, const Mat &, int)
{
throw_nogpu();
return Ptr<BaseRowFilter_GPU>(0);
}
Ptr<BaseColumnFilter_GPU> cv::ocl::getLinearColumnFilter_GPU(int, int, const Mat &, int)
{
throw_nogpu();
return Ptr<BaseColumnFilter_GPU>(0);
}
Ptr<FilterEngine_GPU> cv::ocl::createGaussianFilter_GPU(int, Size, double, double)
{
throw_nogpu();
return Ptr<FilterEngine_GPU>(0);
}
Ptr<BaseFilter_GPU> cv::ocl::getLinearFilter_GPU(int, int, const Mat &, const Size &, Point)
{
throw_nogpu();
return Ptr<BaseFilter_GPU>(0);
}
Ptr<FilterEngine_GPU> cv::ocl::createLinearFilter_GPU(int, int, const Mat &, const Point &)
{
throw_nogpu();
return Ptr<FilterEngine_GPU>(0);
}
Ptr<FilterEngine_GPU> cv::ocl::createDerivFilter_GPU(int srcType, int dstType, int dx, int dy, int ksize, int borderType)
{
throw_nogpu();
return Ptr<FilterEngine_GPU>(0);
}
void cv::ocl::boxFilter(const oclMat &, oclMat &, int, Size, Point, int)
{
throw_nogpu();
}
void cv::ocl::sepFilter2D(const oclMat &, oclMat &, int, const Mat &, const Mat &, Point)
{
throw_nogpu();
}
void cv::ocl::Sobel(const oclMat &, oclMat &, int, int, int, int, double)
{
throw_nogpu();
}
void cv::ocl::Scharr(const oclMat &, oclMat &, int, int, int, double)
{
throw_nogpu();
}
void cv::ocl::GaussianBlur(const oclMat &, oclMat &, Size, double, double)
{
throw_nogpu();
}
void cv::ocl::filter2D(const oclMat &, oclMat &, int, const Mat &, Point)
{
throw_nogpu();
}
void cv::ocl::Laplacian(const oclMat &, oclMat &, int, int, double)
{
throw_nogpu();
}
void cv::ocl::erode(const oclMat &, oclMat &, const Mat &, Point, int)
{
throw_nogpu();
}
void cv::ocl::dilate(const oclMat &, oclMat &, const Mat &, Point, int)
{
throw_nogpu();
}
void cv::ocl::morphologyEx(const oclMat &, oclMat &, int, const Mat &, Point, int)
{
throw_nogpu();
}
#else /* !defined (HAVE_OPENCL) */
//helper routines
namespace cv
{
@ -1705,5 +1590,3 @@ void cv::ocl::GaussianBlur(const oclMat &src, oclMat &dst, Size ksize, double si
Ptr<FilterEngine_GPU> f = createGaussianFilter_GPU(src.type(), ksize, sigma1, sigma2, bordertype);
f->apply(src, dst);
}
#endif /* !defined (HAVE_OPENCL) */

19
modules/ocl/src/gemm.cpp
Unescape View File

@ -46,24 +46,14 @@
#include <iomanip>
#include "precomp.hpp"
#ifdef HAVE_CLAMDBLAS
#include "clAmdBlas.h"
#if !defined HAVE_OPENCL
void cv::ocl::gemm(const oclMat &src1, const oclMat &src2, double alpha,
const oclMat &src3, double beta, oclMat &dst, int flags)
{
throw_nogpu();
}
#elif !defined HAVE_CLAMDBLAS
void cv::ocl::gemm(const oclMat &src1, const oclMat &src2, double alpha,
const oclMat &src3, double beta, oclMat &dst, int flags)
#if !defined HAVE_CLAMDBLAS
void cv::ocl::gemm(const oclMat&, const oclMat&, double,
const oclMat&, double, oclMat&, int)
{
CV_Error(CV_StsNotImplemented, "OpenCL BLAS is not implemented");
}
#else
#include "clAmdBlas.h"
using namespace cv;
void cv::ocl::gemm(const oclMat &src1, const oclMat &src2, double alpha,
@ -168,4 +158,3 @@ void cv::ocl::gemm(const oclMat &src1, const oclMat &src2, double alpha,
clAmdBlasTeardown();
}
#endif
#endif

1256
modules/ocl/src/haar.cpp
View File

File diff suppressed because it is too large Load Diff

65
modules/ocl/src/hog.cpp
Unescape View File

@ -49,69 +49,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
cv::ocl::HOGDescriptor::HOGDescriptor(Size, Size, Size, Size, int, double, double, bool, int)
{
throw_nogpu();
}
size_t cv::ocl::HOGDescriptor::getDescriptorSize() const
{
throw_nogpu();
return 0;
}
size_t cv::ocl::HOGDescriptor::getBlockHistogramSize() const
{
throw_nogpu();
return 0;
}
double cv::ocl::HOGDescriptor::getWinSigma() const
{
throw_nogpu();
return 0;
}
bool cv::ocl::HOGDescriptor::checkDetectorSize() const
{
throw_nogpu();
return false;
}
void cv::ocl::HOGDescriptor::setSVMDetector(const vector<float> &)
{
throw_nogpu();
}
void cv::ocl::HOGDescriptor::detect(const oclMat &, vector<Point> &, double, Size, Size)
{
throw_nogpu();
}
void cv::ocl::HOGDescriptor::detectMultiScale(const oclMat &, vector<Rect> &, double, Size, Size, double, int)
{
throw_nogpu();
}
void cv::ocl::HOGDescriptor::computeBlockHistograms(const oclMat &)
{
throw_nogpu();
}
void cv::ocl::HOGDescriptor::getDescriptors(const oclMat &, Size, oclMat &, int)
{
throw_nogpu();
}
std::vector<float> cv::ocl::HOGDescriptor::getDefaultPeopleDetector()
{
throw_nogpu();
return std::vector<float>();
}
std::vector<float> cv::ocl::HOGDescriptor::getPeopleDetector48x96()
{
throw_nogpu();
return std::vector<float>();
}
std::vector<float> cv::ocl::HOGDescriptor::getPeopleDetector64x128()
{
throw_nogpu();
return std::vector<float>();
}
#else
#define CELL_WIDTH 8
#define CELL_HEIGHT 8
@ -1895,5 +1832,3 @@ void cv::ocl::device::hog::resize( const oclMat &src, oclMat &dst, const Size sz
openCLExecuteKernel2(clCxt, &objdetect_hog, kernelName, globalThreads, localThreads, args, -1, -1);
}
#endif

57
modules/ocl/src/imgproc.cpp
Unescape View File

@ -59,62 +59,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
void cv::ocl::meanShiftFiltering(const oclMat &, oclMat &, int, int, TermCriteria)
{
throw_nogpu();
}
void cv::ocl::meanShiftProc(const oclMat &, oclMat &, oclMat &, int, int, TermCriteria)
{
throw_nogpu();
}
double cv::ocl::threshold(const oclMat &, oclMat &, double, int)
{
throw_nogpu();
return 0.0;
}
void cv::ocl::resize(const oclMat &, oclMat &, Size, double, double, int)
{
throw_nogpu();
}
void cv::ocl::remap(const oclMat &, oclMat &, oclMat &, oclMat &, int, int , const Scalar &)
{
throw_nogpu();
}
void cv::ocl::copyMakeBorder(const oclMat &, oclMat &, int, int, int, int, const Scalar &)
{
throw_nogpu();
}
void cv::ocl::warpAffine(const oclMat &, oclMat &, const Mat &, Size, int)
{
throw_nogpu();
}
void cv::ocl::warpPerspective(const oclMat &, oclMat &, const Mat &, Size, int)
{
throw_nogpu();
}
void cv::ocl::integral(const oclMat &, oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::calcHist(const oclMat &, oclMat &hist)
{
throw_nogpu();
}
void cv::ocl::bilateralFilter(const oclMat &, oclMat &, int, double, double, int)
{
throw_nogpu();
}
void cv::ocl::convolve(const oclMat &, const oclMat &, oclMat &)
{
throw_nogpu();
}
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -1696,4 +1640,3 @@ void cv::ocl::convolve(const oclMat &x, const oclMat &t, oclMat &y)
convolve_run(x, t, y, kernelName, &imgproc_convolve);
}
#endif /* !defined (HAVE_OPENCL) */

77
modules/ocl/src/initialization.cpp
Unescape View File

@ -61,82 +61,6 @@ using std::endl;
//#define AMD_DOUBLE_DIFFER
#if !defined (HAVE_OPENCL)
namespace cv
{
namespace ocl
{
cl_device_id getDevice()
{
throw_nogpu();
return 0;
}
void getComputeCapability(cl_device_id, int &major, int &minor)
{
throw_nogpu();
}
void openCLMallocPitch(Context * /*clCxt*/, void ** /*dev_ptr*/, size_t * /*pitch*/,
size_t /*widthInBytes*/, size_t /*height*/)
{
throw_nogpu();
}
void openCLMemcpy2D(Context * /*clCxt*/, void * /*dst*/, size_t /*dpitch*/,
const void * /*src*/, size_t /*spitch*/,
size_t /*width*/, size_t /*height*/, enum openCLMemcpyKind /*kind*/)
{
throw_nogpu();
}
void openCLCopyBuffer2D(Context * /*clCxt*/, void * /*dst*/, size_t /*dpitch*/,
const void * /*src*/, size_t /*spitch*/,
size_t /*width*/, size_t /*height*/, enum openCLMemcpyKind /*kind*/)
{
throw_nogpu();
}
cl_mem openCLCreateBuffer(Context *, size_t, size_t)
{
throw_nogpu();
}
void openCLReadBuffer(Context *, cl_mem, void *, size_t)
{
throw_nogpu();
}
void openCLFree(void * /*devPtr*/)
{
throw_nogpu();
}
cl_kernel openCLGetKernelFromSource(const Context * /*clCxt*/,
const char ** /*fileName*/, string /*kernelName*/)
{
throw_nogpu();
}
void openCLVerifyKernel(const Context * /*clCxt*/, cl_kernel /*kernel*/, size_t * /*blockSize*/,
size_t * /*globalThreads*/, size_t * /*localThreads*/)
{
throw_nogpu();
}
cl_mem load_constant(cl_context context, cl_command_queue command_queue, const void *value,
const size_t size)
{
throw_nogpu();
}
}//namespace ocl
}//namespace cv
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -1013,4 +937,3 @@ namespace cv
}//namespace ocl
}//namespace cv
#endif

12
modules/ocl/src/interpolate_frames.cpp
Unescape View File

@ -50,17 +50,6 @@ using namespace std;
using namespace cv;
using namespace cv::ocl;
#if !defined (HAVE_OPENCL)
void cv::ocl::interpolateFrames(const oclMat &frame0, const oclMat &frame1,
const oclMat &fu, const oclMat &fv,
const oclMat &bu, const oclMat &bv,
float pos, oclMat &newFrame, oclMat &buf)
{
throw_nogpu();
}
#else
namespace cv
{
namespace ocl
@ -311,5 +300,4 @@ void interpolate::bindImgTex(const oclMat &img, cl_mem &texture)
clEnqueueCopyBufferToImage(img.clCxt->impl->clCmdQueue, (cl_mem)img.data, texture, 0, origin, region, 0, NULL, 0);
openCLSafeCall(err);
}
#endif//(HAVE_OPENCL)

409
modules/ocl/src/kernels/haarobjectdetect.cl
Unescape View File

@ -9,6 +9,7 @@
// Niko Li, newlife20080214@gmail.com
// Wang Weiyan, wangweiyanster@gmail.com
// Jia Haipeng, jiahaipeng95@gmail.com
// Nathan, liujun@multicorewareinc.com
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@ -47,14 +48,14 @@ typedef float sqsumtype;
typedef struct __attribute__((aligned (128))) GpuHidHaarFeature
{
struct __attribute__((aligned (32)))
{
int p0 __attribute__((aligned (4)));
int p1 __attribute__((aligned (4)));
int p2 __attribute__((aligned (4)));
int p3 __attribute__((aligned (4)));
float weight __attribute__((aligned (4)));
}
rect[CV_HAAR_FEATURE_MAX] __attribute__((aligned (32)));
{
int p0 __attribute__((aligned (4)));
int p1 __attribute__((aligned (4)));
int p2 __attribute__((aligned (4)));
int p3 __attribute__((aligned (4)));
float weight __attribute__((aligned (4)));
}
rect[CV_HAAR_FEATURE_MAX] __attribute__((aligned (32)));
}
GpuHidHaarFeature;
@ -108,31 +109,31 @@ typedef struct __attribute__((aligned (64))) GpuHidHaarClassifierCascade
int p2 __attribute__((aligned (4)));
int p3 __attribute__((aligned (4)));
float inv_window_area __attribute__((aligned (4)));
}GpuHidHaarClassifierCascade;
} GpuHidHaarClassifierCascade;
__kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCascade(//constant GpuHidHaarClassifierCascade * cascade,
global GpuHidHaarStageClassifier * stagecascadeptr,
global int4 * info,
global GpuHidHaarTreeNode * nodeptr,
global const int * restrict sum1,
global const float * restrict sqsum1,
global int4 * candidate,
const int pixelstep,
const int loopcount,
const int start_stage,
const int split_stage,
const int end_stage,
const int startnode,
const int splitnode,
const int4 p,
const int4 pq,
const float correction
//const int width,
//const int height,
//const int grpnumperline,
//const int totalgrp
)
global GpuHidHaarStageClassifier * stagecascadeptr,
global int4 * info,
global GpuHidHaarTreeNode * nodeptr,
global const int * restrict sum1,
global const float * restrict sqsum1,
global int4 * candidate,
const int pixelstep,
const int loopcount,
const int start_stage,
const int split_stage,
const int end_stage,
const int startnode,
const int splitnode,
const int4 p,
const int4 pq,
const float correction
//const int width,
//const int height,
//const int grpnumperline,
//const int totalgrp
)
{
int grpszx = get_local_size(0);
int grpszy = get_local_size(1);
@ -184,7 +185,7 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
__global const int * sum = sum1 + imgoff;
__global const float * sqsum = sqsum1 + imgoff;
for(int grploop=grpidx;grploop<totalgrp;grploop+=grpnumx)
for(int grploop=grpidx; grploop<totalgrp; grploop+=grpnumx)
{
int grpidy = grploop / grpnumperline;
int grpidx = grploop - mul24(grpidy, grpnumperline);
@ -195,7 +196,7 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int grpoffx = x-lclidx;
int grpoffy = y-lclidy;
for(int i=0;i<read_loop;i++)
for(int i=0; i<read_loop; i++)
{
int pos_id = mad24(i,lcl_sz,lcl_id);
pos_id = pos_id < total_read ? pos_id : 0;
@ -234,15 +235,15 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
cascadeinfo1.x +=lcl_off;
cascadeinfo1.z +=lcl_off;
mean = (lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.x)] - lcldata[mad24(cascadeinfo1.y,readwidth,cascadeinfo1.z)] -
lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.x)] + lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.z)])
*correction;
lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.x)] + lcldata[mad24(cascadeinfo1.w,readwidth,cascadeinfo1.z)])
*correction;
int p_offset = mad24(y, pixelstep, x);
cascadeinfo2.x +=p_offset;
cascadeinfo2.z +=p_offset;
variance_norm_factor =sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.x)] - sqsum[mad24(cascadeinfo2.y, pixelstep, cascadeinfo2.z)] -
sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.x)] + sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.z)];
sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.x)] + sqsum[mad24(cascadeinfo2.w, pixelstep, cascadeinfo2.z)];
variance_norm_factor = variance_norm_factor * correction - mean * mean;
variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1.f;
@ -270,19 +271,19 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
info2.z +=lcl_off;
float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
//if((info3.z - info3.x) && (!stageinfo.z))
//{
info3.x +=lcl_off;
info3.z +=lcl_off;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
info3.x +=lcl_off;
info3.z +=lcl_off;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
//}
stage_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
nodecounter++;
@ -299,12 +300,13 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
}
barrier(CLK_LOCAL_MEM_FENCE);
int queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE);
nodecounter = splitnode;
for(int stageloop = split_stage; stageloop< end_stage && queuecount>0;stageloop++)
for(int stageloop = split_stage; stageloop< end_stage && queuecount>0; stageloop++)
{
//barrier(CLK_LOCAL_MEM_FENCE);
//barrier(CLK_LOCAL_MEM_FENCE);
//if(lcl_id == 0)
lclcount[0]=0;
lclcount[0]=0;
barrier(CLK_LOCAL_MEM_FENCE);
int2 stageinfo = *(global int2*)(stagecascadeptr+stageloop);
@ -316,70 +318,73 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
int lcl_compute_win_id = (lcl_id >>(6-perfscale));
int lcl_loops = (stageinfo.x + lcl_compute_win -1) >> (6-perfscale);
int lcl_compute_id = lcl_id - (lcl_compute_win_id << (6-perfscale));
for(int queueloop=0;queueloop<queuecount_loop/* && lcl_compute_win_id < queuecount*/;queueloop++)
for(int queueloop=0; queueloop<queuecount_loop/* && lcl_compute_win_id < queuecount*/; queueloop++)
{
float stage_sum = 0.f;
int temp_coord = lcloutindex[lcl_compute_win_id<<1];
float variance_norm_factor = as_float(lcloutindex[(lcl_compute_win_id<<1)+1]);
int queue_pixel = mad24(((temp_coord & (int)0xffff0000)>>16),readwidth,temp_coord & 0xffff);
//barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount) {
int tempnodecounter = lcl_compute_id;
float part_sum = 0.f;
for(int lcl_loop=0;lcl_loop<lcl_loops && tempnodecounter<stageinfo.x;lcl_loop++)
//barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount)
{
__global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter + tempnodecounter);
int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
float nodethreshold = w.w * variance_norm_factor;
int tempnodecounter = lcl_compute_id;
float part_sum = 0.f;
for(int lcl_loop=0; lcl_loop<lcl_loops && tempnodecounter<stageinfo.x; lcl_loop++)
{
__global GpuHidHaarTreeNode* currentnodeptr = (nodeptr + nodecounter + tempnodecounter);
info1.x +=queue_pixel;
info1.z +=queue_pixel;
info2.x +=queue_pixel;
info2.z +=queue_pixel;
int4 info1 = *(__global int4*)(&(currentnodeptr->p[0][0]));
int4 info2 = *(__global int4*)(&(currentnodeptr->p[1][0]));
int4 info3 = *(__global int4*)(&(currentnodeptr->p[2][0]));
float4 w = *(__global float4*)(&(currentnodeptr->weight[0]));
float2 alpha2 = *(__global float2*)(&(currentnodeptr->alpha[0]));
float nodethreshold = w.w * variance_norm_factor;
float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
info1.x +=queue_pixel;
info1.z +=queue_pixel;
info2.x +=queue_pixel;
info2.z +=queue_pixel;
float classsum = (lcldata[mad24(info1.y,readwidth,info1.x)] - lcldata[mad24(info1.y,readwidth,info1.z)] -
lcldata[mad24(info1.w,readwidth,info1.x)] + lcldata[mad24(info1.w,readwidth,info1.z)]) * w.x;
classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
//if((info3.z - info3.x) && (!stageinfo.z))
//{
classsum += (lcldata[mad24(info2.y,readwidth,info2.x)] - lcldata[mad24(info2.y,readwidth,info2.z)] -
lcldata[mad24(info2.w,readwidth,info2.x)] + lcldata[mad24(info2.w,readwidth,info2.z)]) * w.y;
//if((info3.z - info3.x) && (!stageinfo.z))
//{
info3.x +=queue_pixel;
info3.z +=queue_pixel;
classsum += (lcldata[mad24(info3.y,readwidth,info3.x)] - lcldata[mad24(info3.y,readwidth,info3.z)] -
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
//}
part_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
tempnodecounter +=lcl_compute_win;
}//end for(int lcl_loop=0;lcl_loop<lcl_loops;lcl_loop++)
partialsum[lcl_id]=part_sum;
}
barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount) {
for(int i=0;i<lcl_compute_win && (lcl_compute_id==0);i++)
{
stage_sum += partialsum[lcl_id+i];
lcldata[mad24(info3.w,readwidth,info3.x)] + lcldata[mad24(info3.w,readwidth,info3.z)]) * w.z;
//}
part_sum += classsum >= nodethreshold ? alpha2.y : alpha2.x;
tempnodecounter +=lcl_compute_win;
}//end for(int lcl_loop=0;lcl_loop<lcl_loops;lcl_loop++)
partialsum[lcl_id]=part_sum;
}
if(stage_sum >= stagethreshold && (lcl_compute_id==0))
barrier(CLK_LOCAL_MEM_FENCE);
if(lcl_compute_win_id < queuecount)
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex<<1] = temp_coord;
lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
for(int i=0; i<lcl_compute_win && (lcl_compute_id==0); i++)
{
stage_sum += partialsum[lcl_id+i];
}
if(stage_sum >= stagethreshold && (lcl_compute_id==0))
{
int queueindex = atomic_inc(lclcount);
lcloutindex[queueindex<<1] = temp_coord;
lcloutindex[(queueindex<<1)+1] = as_int(variance_norm_factor);
}
lcl_compute_win_id +=(1<<perfscale);
}
lcl_compute_win_id +=(1<<perfscale);
}
barrier(CLK_LOCAL_MEM_FENCE);
}//end for(int queueloop=0;queueloop<queuecount_loop;queueloop++)
barrier(CLK_LOCAL_MEM_FENCE);
//barrier(CLK_LOCAL_MEM_FENCE);
queuecount = lclcount[0];
barrier(CLK_LOCAL_MEM_FENCE);
nodecounter += stageinfo.x;
}//end for(int stageloop = splitstage; stageloop< endstage && queuecount>0;stageloop++)
//barrier(CLK_LOCAL_MEM_FENCE);
@ -420,138 +425,138 @@ __kernel void __attribute__((reqd_work_group_size(8,8,1)))gpuRunHaarClassifierCa
/*
if(stagecascade->two_rects)
{
#pragma unroll
for( n = 0; n < stagecascade->count; n++ )
{
t1 = *(node + counter);
t = t1.threshold * variance_norm_factor;
classsum = calc_sum1(t1,p_offset,0) * t1.weight[0];
/*
if(stagecascade->two_rects)
{
#pragma unroll
for( n = 0; n < stagecascade->count; n++ )
{
t1 = *(node + counter);
t = t1.threshold * variance_norm_factor;
classsum = calc_sum1(t1,p_offset,0) * t1.weight[0];
classsum += calc_sum1(t1, p_offset,1) * t1.weight[1];
stage_sum += classsum >= t ? t1.alpha[1]:t1.alpha[0];
classsum += calc_sum1(t1, p_offset,1) * t1.weight[1];
stage_sum += classsum >= t ? t1.alpha[1]:t1.alpha[0];
counter++;
}
}
else
{
#pragma unroll
for( n = 0; n < stagecascade->count; n++ )
{
t = node[counter].threshold*variance_norm_factor;
classsum = calc_sum1(node[counter],p_offset,0) * node[counter].weight[0];
classsum += calc_sum1(node[counter],p_offset,1) * node[counter].weight[1];
counter++;
}
}
else
{
#pragma unroll
for( n = 0; n < stagecascade->count; n++ )
{
t = node[counter].threshold*variance_norm_factor;
classsum = calc_sum1(node[counter],p_offset,0) * node[counter].weight[0];
classsum += calc_sum1(node[counter],p_offset,1) * node[counter].weight[1];
if( node[counter].p0[2] )
classsum += calc_sum1(node[counter],p_offset,2) * node[counter].weight[2];
if( node[counter].p0[2] )
classsum += calc_sum1(node[counter],p_offset,2) * node[counter].weight[2];
stage_sum += classsum >= t ? node[counter].alpha[1]:node[counter].alpha[0];// modify
stage_sum += classsum >= t ? node[counter].alpha[1]:node[counter].alpha[0];// modify
counter++;
}
}
*/
/*
counter++;
}
}
*/
/*
__kernel void gpuRunHaarClassifierCascade_ScaleWindow(
constant GpuHidHaarClassifierCascade * _cascade,
global GpuHidHaarStageClassifier * stagecascadeptr,
//global GpuHidHaarClassifier * classifierptr,
global GpuHidHaarTreeNode * nodeptr,
global int * sum,
global float * sqsum,
global int * _candidate,
int pixel_step,
int cols,
int rows,
int start_stage,
int end_stage,
//int counts,
int nodenum,
int ystep,
int detect_width,
//int detect_height,
int loopcount,
int outputstep)
//float scalefactor)
constant GpuHidHaarClassifierCascade * _cascade,
global GpuHidHaarStageClassifier * stagecascadeptr,
//global GpuHidHaarClassifier * classifierptr,
global GpuHidHaarTreeNode * nodeptr,
global int * sum,
global float * sqsum,
global int * _candidate,
int pixel_step,
int cols,
int rows,
int start_stage,
int end_stage,
//int counts,
int nodenum,
int ystep,
int detect_width,
//int detect_height,
int loopcount,
int outputstep)
//float scalefactor)
{
unsigned int x1 = get_global_id(0);
unsigned int y1 = get_global_id(1);
int p_offset;
int m, n;
int result;
int counter;
float mean, variance_norm_factor;
for(int i=0;i<loopcount;i++)
{
constant GpuHidHaarClassifierCascade * cascade = _cascade + i;
global int * candidate = _candidate + i*outputstep;
int window_width = cascade->p1 - cascade->p0;
int window_height = window_width;
result = 1;
counter = 0;
unsigned int x = mul24(x1,ystep);
unsigned int y = mul24(y1,ystep);
if((x < cols - window_width - 1) && (y < rows - window_height -1))
{
global GpuHidHaarStageClassifier *stagecascade = stagecascadeptr +cascade->count*i+ start_stage;
//global GpuHidHaarClassifier *classifier = classifierptr;
global GpuHidHaarTreeNode *node = nodeptr + nodenum*i;
unsigned int x1 = get_global_id(0);
unsigned int y1 = get_global_id(1);
int p_offset;
int m, n;
int result;
int counter;
float mean, variance_norm_factor;
for(int i=0;i<loopcount;i++)
{
constant GpuHidHaarClassifierCascade * cascade = _cascade + i;
global int * candidate = _candidate + i*outputstep;
int window_width = cascade->p1 - cascade->p0;
int window_height = window_width;
result = 1;
counter = 0;
unsigned int x = mul24(x1,ystep);
unsigned int y = mul24(y1,ystep);
if((x < cols - window_width - 1) && (y < rows - window_height -1))
{
global GpuHidHaarStageClassifier *stagecascade = stagecascadeptr +cascade->count*i+ start_stage;
//global GpuHidHaarClassifier *classifier = classifierptr;
global GpuHidHaarTreeNode *node = nodeptr + nodenum*i;
p_offset = mad24(y, pixel_step, x);// modify
p_offset = mad24(y, pixel_step, x);// modify
mean = (*(sum + p_offset + (int)cascade->p0) - *(sum + p_offset + (int)cascade->p1) -
*(sum + p_offset + (int)cascade->p2) + *(sum + p_offset + (int)cascade->p3))
*cascade->inv_window_area;
mean = (*(sum + p_offset + (int)cascade->p0) - *(sum + p_offset + (int)cascade->p1) -
*(sum + p_offset + (int)cascade->p2) + *(sum + p_offset + (int)cascade->p3))
*cascade->inv_window_area;
variance_norm_factor = *(sqsum + p_offset + cascade->p0) - *(sqsum + cascade->p1 + p_offset) -
*(sqsum + p_offset + cascade->p2) + *(sqsum + cascade->p3 + p_offset);
variance_norm_factor = variance_norm_factor * cascade->inv_window_area - mean * mean;
variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1;//modify
variance_norm_factor = *(sqsum + p_offset + cascade->p0) - *(sqsum + cascade->p1 + p_offset) -
*(sqsum + p_offset + cascade->p2) + *(sqsum + cascade->p3 + p_offset);
variance_norm_factor = variance_norm_factor * cascade->inv_window_area - mean * mean;
variance_norm_factor = variance_norm_factor >=0.f ? sqrt(variance_norm_factor) : 1;//modify
// if( cascade->is_stump_based )
//{
for( m = start_stage; m < end_stage; m++ )
{
float stage_sum = 0.f;
float t, classsum;
GpuHidHaarTreeNode t1;
// if( cascade->is_stump_based )
//{
for( m = start_stage; m < end_stage; m++ )
{
float stage_sum = 0.f;
float t, classsum;
GpuHidHaarTreeNode t1;
//#pragma unroll
for( n = 0; n < stagecascade->count; n++ )
{
t1 = *(node + counter);
t = t1.threshold * variance_norm_factor;
classsum = calc_sum1(t1, p_offset ,0) * t1.weight[0] + calc_sum1(t1, p_offset ,1) * t1.weight[1];
//#pragma unroll
for( n = 0; n < stagecascade->count; n++ )
{
t1 = *(node + counter);
t = t1.threshold * variance_norm_factor;
classsum = calc_sum1(t1, p_offset ,0) * t1.weight[0] + calc_sum1(t1, p_offset ,1) * t1.weight[1];
if((t1.p0[2]) && (!stagecascade->two_rects))
classsum += calc_sum1(t1, p_offset, 2) * t1.weight[2];
if((t1.p0[2]) && (!stagecascade->two_rects))
classsum += calc_sum1(t1, p_offset, 2) * t1.weight[2];
stage_sum += classsum >= t ? t1.alpha[1] : t1.alpha[0];// modify
counter++;
}
stage_sum += classsum >= t ? t1.alpha[1] : t1.alpha[0];// modify
counter++;
}
if (stage_sum < stagecascade->threshold)
{
result = 0;
break;
}
if (stage_sum < stagecascade->threshold)
{
result = 0;
break;
}
stagecascade++;
stagecascade++;
}
if(result)
{
candidate[4 * (y1 * detect_width + x1)] = x;
candidate[4 * (y1 * detect_width + x1) + 1] = y;
candidate[4 * (y1 * detect_width + x1)+2] = window_width;
candidate[4 * (y1 * detect_width + x1) + 3] = window_height;
}
//}
}
}
}
if(result)
{
candidate[4 * (y1 * detect_width + x1)] = x;
candidate[4 * (y1 * detect_width + x1) + 1] = y;
candidate[4 * (y1 * detect_width + x1)+2] = window_width;
candidate[4 * (y1 * detect_width + x1) + 3] = window_height;
}
//}
}
}
}
*/

7
modules/ocl/src/match_template.cpp
Unescape View File

@ -51,12 +51,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
void cv::ocl::matchTemplate(const oclMat &, const oclMat &, oclMat &)
{
throw_nogpu();
}
#else
//helper routines
namespace cv
{
@ -498,4 +492,3 @@ void cv::ocl::matchTemplate(const oclMat &image, const oclMat &templ, oclMat &re
CV_Assert(caller);
caller(image, templ, result, buf);
}
#endif //

55
modules/ocl/src/matrix_operations.cpp
Unescape View File

@ -58,60 +58,6 @@ using namespace std;
//////////////////////////////// oclMat ////////////////////////////////
////////////////////////////////////////////////////////////////////////
#if !defined (HAVE_OPENCL)
namespace cv
{
namespace ocl
{
void oclMat::upload(const Mat & /*m*/)
{
throw_nogpu();
}
void oclMat::download(cv::Mat & /*m*/) const
{
throw_nogpu();
}
void oclMat::copyTo( oclMat & /*m*/ ) const
{
throw_nogpu();
}
void oclMat::copyTo( oclMat & /*m*/, const oclMat &/* mask */) const
{
throw_nogpu();
}
void oclMat::convertTo( oclMat & /*m*/, int /*rtype*/, double /*alpha*/, double /*beta*/ ) const
{
throw_nogpu();
}
oclMat &oclMat::operator = (const Scalar & /*s*/)
{
throw_nogpu();
return *this;
}
oclMat &oclMat::setTo(const Scalar & /*s*/, const oclMat & /*mask*/)
{
throw_nogpu();
return *this;
}
oclMat oclMat::reshape(int /*new_cn*/, int /*new_rows*/) const
{
throw_nogpu();
return oclMat();
}
void oclMat::create(int /*_rows*/, int /*_cols*/, int /*_type*/)
{
throw_nogpu();
}
void oclMat::release()
{
throw_nogpu();
}
}
}
#else /* !defined (HAVE_OPENCL) */
//helper routines
namespace cv
{
@ -1045,4 +991,3 @@ oclMat& cv::ocl::oclMat::operator/=( const oclMat& m )
divide(*this, m, *this);
return *this;
}
#endif /* !defined (HAVE_OPENCL) */

5
modules/ocl/src/mcwutil.hpp
Unescape View File

@ -47,9 +47,6 @@
#define _OPENCV_MCWUTIL_
#include "precomp.hpp"
#if defined (HAVE_OPENCL)
using namespace std;
namespace cv
@ -76,5 +73,5 @@ namespace cv
}//namespace ocl
}//namespace cv
#endif // HAVE_OPENCL
#endif //_OPENCV_MCWUTIL_

18
modules/ocl/src/mssegmentation.cpp
Unescape View File

@ -44,23 +44,6 @@
#include "precomp.hpp"
#if !defined(HAVE_OPENCL)
namespace cv
{
namespace ocl
{
void meanShiftSegmentation(const oclMat &, Mat &, int, int, int, TermCriteria)
{
throw_nogpu();
}
}
}
#else
using namespace std;
// Auxiliray stuff
@ -411,4 +394,3 @@ namespace cv
}
}
#endif // #if !defined (HAVE_OPENCL)

9
modules/ocl/src/pyrlk.cpp
Unescape View File

@ -52,13 +52,6 @@ using namespace std;
using namespace cv;
using namespace cv::ocl;
#if !defined (HAVE_OPENCL)
void cv::ocl::PyrLKOpticalFlow::sparse(const oclMat &, const oclMat &, const oclMat &, oclMat &, oclMat &, oclMat &) { }
void cv::ocl::PyrLKOpticalFlow::dense(const oclMat &, const oclMat &, oclMat &, oclMat &, oclMat *) { }
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -862,5 +855,3 @@ void cv::ocl::PyrLKOpticalFlow::dense(const oclMat &prevImg, const oclMat &nextI
clFinish(prevImg.clCxt->impl->clCmdQueue);
}
#endif /* !defined (HAVE_CUDA) */

10
modules/ocl/src/pyrup.cpp
Unescape View File

@ -54,13 +54,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#ifndef HAVE_OPENCL
void cv::ocl::pyrUp(const oclMat &, GpuMat &, oclMat &)
{
throw_nogpu();
}
#else
namespace cv
{
namespace ocl
@ -93,5 +86,4 @@ namespace cv
openCLExecuteKernel(clCxt, &pyr_up, kernelName, globalThreads, localThreads, args, src.oclchannels(), src.depth());
}
}
};
#endif // HAVE_OPENCL
}

29
modules/ocl/src/split_merge.cpp
Unescape View File

@ -58,34 +58,6 @@ using std::endl;
///////////////// oclMat merge and split ///////////////////////////////
////////////////////////////////////////////////////////////////////////
#if !defined (HAVE_OPENCL)
namespace cv
{
namespace ocl
{
void cv::ocl::merge(const oclMat *src_mat, size_t count, oclMat &dst_mat)
{
throw_nogpu();
}
void cv::ocl::merge(const vector<oclMat> &src_mat, oclMat &dst_mat)
{
throw_nogpu();
}
void cv::ocl::split(const oclMat &src, oclMat *dst)
{
throw_nogpu();
}
void cv::ocl::split(const oclMat &src, vector<oclMat> &dst)
{
throw_nogpu();
}
}
}
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -417,4 +389,3 @@ void cv::ocl::split(const oclMat &src, vector<oclMat> &dst)
if(src.oclchannels() > 0)
split_merge::split(src, &dst[0]);
}
#endif /* !defined (HAVE_OPENCL) */

55
modules/ocl/src/surf.cpp
Unescape View File

@ -50,59 +50,6 @@ using namespace cv;
using namespace cv::ocl;
using namespace std;
#if !defined (HAVE_OPENCL)
cv::ocl::SURF_OCL::SURF_OCL()
{
throw_nogpu();
}
cv::ocl::SURF_OCL::SURF_OCL(double, int, int, bool, float, bool)
{
throw_nogpu();
}
int cv::ocl::SURF_OCL::descriptorSize() const
{
throw_nogpu();
return 0;
}
void cv::ocl::SURF_OCL::uploadKeypoints(const vector<KeyPoint> &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::downloadKeypoints(const oclMat &, vector<KeyPoint> &)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::downloadDescriptors(const oclMat &, vector<float> &)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::operator()(const oclMat &, const oclMat &, oclMat &)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::operator()(const oclMat &, const oclMat &, oclMat &, oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::operator()(const oclMat &, const oclMat &, vector<KeyPoint> &)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::operator()(const oclMat &, const oclMat &, vector<KeyPoint> &, oclMat &, bool)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::operator()(const oclMat &, const oclMat &, vector<KeyPoint> &, vector<float> &, bool)
{
throw_nogpu();
}
void cv::ocl::SURF_OCL::releaseMemory()
{
throw_nogpu();
}
#else /* !defined (HAVE_OPENCL) */
namespace cv
{
namespace ocl
@ -755,5 +702,3 @@ void SURF_OCL_Invoker::compute_descriptors_gpu(const oclMat &descriptors, const
}
}
#endif // /* !defined (HAVE_OPENCL) */

4
modules/ocl/test/test_imgproc.cpp
Unescape View File

@ -791,8 +791,8 @@ TEST_P(WarpAffine, Mat)
{
static const double coeffs[2][3] =
{
{cos(3.14 / 6), -sin(3.14 / 6), 100.0},
{sin(3.14 / 6), cos(3.14 / 6), -100.0}
{cos(CV_PI / 6), -sin(CV_PI / 6), 100.0},
{sin(CV_PI / 6), cos(CV_PI / 6), -100.0}
};
Mat M(2, 3, CV_64F, (void *)coeffs);

21
modules/ocl/test/test_match_template.cpp
Unescape View File

@ -44,7 +44,7 @@
#include "precomp.hpp"
//#define PERF_TEST 0
#ifdef HAVE_OPENCL
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate
@ -97,15 +97,7 @@ TEST_P(MatchTemplate8U, Accuracy)
cv::Mat mat_dst;
dst.download(mat_dst);
EXPECT_MAT_NEAR(dst_gold, mat_dst, templ_size.area() * 1e-1, sss);
#ifdef PERF_TEST
{
P_TEST_FULL( {}, {cv::ocl::matchTemplate(ocl_image, ocl_templ, dst, method);}, {});
P_TEST_FULL( {}, {cv::matchTemplate(image, templ, dst_gold, method);}, {});
}
#endif // PERF_TEST
}
PARAM_TEST_CASE(MatchTemplate32F, cv::Size, TemplateSize, Channels, TemplateMethod)
@ -144,17 +136,6 @@ TEST_P(MatchTemplate32F, Accuracy)
dst.download(mat_dst);
EXPECT_MAT_NEAR(dst_gold, mat_dst, templ_size.area() * 1e-1, sss);
#ifdef PERF_TEST
{
std::cout << "Method: " << TEMPLATE_METHOD_NAMES[method] << std::endl;
std::cout << "Image Size: (" << size.width << ", " << size.height << ")" << std::endl;
std::cout << "Template Size: (" << templ_size.width << ", " << templ_size.height << ")" << std::endl;
std::cout << "Channels: " << cn << std::endl;
P_TEST_FULL( {}, {cv::ocl::matchTemplate(ocl_image, ocl_templ, dst, method);}, {});
P_TEST_FULL( {}, {cv::matchTemplate(image, templ, dst_gold, method);}, {});
}
#endif // PERF_TEST
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate8U,

Write Preview
Loading…
Cancel
Save