generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fixed gpu::filter2D border interpolation for CV_32FC1 type

Browse Source 
added additional tests for gpu filters
fixed gpu features2D tests
This commit is contained in:
Vladislav Vinogradov
2012-03-21 14:38:23 +00:00
parent c1a6cb6221
commit 059cef57e6
16 changed files with 1730 additions and 1515 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
modules/gpu/src/cuda/imgproc.cu
View File

@@ -46,16 +46,16 @@
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/border_interpolate.hpp"
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace imgproc
namespace imgproc
{
/////////////////////////////////// MeanShiftfiltering ///////////////////////////////////////////////
texture<uchar4, 2> tex_meanshift;
__device__ short2 do_mean_shift(int x0, int y0, unsigned char* out,
size_t out_step, int cols, int rows,
__device__ short2 do_mean_shift(int x0, int y0, unsigned char* out,
size_t out_step, int cols, int rows,
int sp, int sr, int maxIter, float eps)
{
int isr2 = sr*sr;
@@ -78,7 +78,7 @@ namespace cv { namespace gpu { namespace device
{
int rowCount = 0;
for( int x = minx; x <= maxx; x++ )
{
{
uchar4 t = tex2D( tex_meanshift, x, y );
int norm2 = (t.x - c.x) * (t.x - c.x) + (t.y - c.y) * (t.y - c.y) + (t.z - c.z) * (t.z - c.z);
@@ -128,16 +128,16 @@ namespace cv { namespace gpu { namespace device
do_mean_shift(x0, y0, out, out_step, cols, rows, sp, sr, maxIter, eps);
}
__global__ void meanshiftproc_kernel(unsigned char* outr, size_t outrstep,
unsigned char* outsp, size_t outspstep,
int cols, int rows,
__global__ void meanshiftproc_kernel(unsigned char* outr, size_t outrstep,
unsigned char* outsp, size_t outspstep,
int cols, int rows,
int sp, int sr, int maxIter, float eps)
{
int x0 = blockIdx.x * blockDim.x + threadIdx.x;
int y0 = blockIdx.y * blockDim.y + threadIdx.y;
if( x0 < cols && y0 < rows )
{
{
int basesp = (blockIdx.y * blockDim.y + threadIdx.y) * outspstep + (blockIdx.x * blockDim.x + threadIdx.x) * 2 * sizeof(short);
*(short2*)(outsp + basesp) = do_mean_shift(x0, y0, outr, outrstep, cols, rows, sp, sr, maxIter, eps);
}
@@ -159,10 +159,10 @@ namespace cv { namespace gpu { namespace device
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
//cudaSafeCall( cudaUnbindTexture( tex_meanshift ) );
//cudaSafeCall( cudaUnbindTexture( tex_meanshift ) );
}
void meanShiftProc_gpu(const DevMem2Db& src, DevMem2Db dstr, DevMem2Db dstsp, int sp, int sr, int maxIter, float eps, cudaStream_t stream)
void meanShiftProc_gpu(const DevMem2Db& src, DevMem2Db dstr, DevMem2Db dstsp, int sp, int sr, int maxIter, float eps, cudaStream_t stream)
{
dim3 grid(1, 1, 1);
dim3 threads(32, 8, 1);
@@ -178,14 +178,14 @@ namespace cv { namespace gpu { namespace device
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
//cudaSafeCall( cudaUnbindTexture( tex_meanshift ) );
//cudaSafeCall( cudaUnbindTexture( tex_meanshift ) );
}
/////////////////////////////////// drawColorDisp ///////////////////////////////////////////////
template <typename T>
__device__ unsigned int cvtPixel(T d, int ndisp, float S = 1, float V = 1)
{
{
unsigned int H = ((ndisp-d) * 240)/ndisp;
unsigned int hi = (H/60) % 6;
@@ -195,7 +195,7 @@ namespace cv { namespace gpu { namespace device
float t = V * (1 - (1 - f) * S);
float3 res;
if (hi == 0) //R = V, G = t, B = p
{
res.x = p;
@@ -208,15 +208,15 @@ namespace cv { namespace gpu { namespace device
res.x = p;
res.y = V;
res.z = q;
}
}
if (hi == 2) // R = p, G = V, B = t
{
res.x = t;
res.y = V;
res.z = p;
}
if (hi == 3) // R = p, G = q, B = V
{
res.x = V;
@@ -242,15 +242,15 @@ namespace cv { namespace gpu { namespace device
const unsigned int r = (unsigned int)(::max(0.f, ::min(res.z, 1.f)) * 255.f);
const unsigned int a = 255U;
return (a << 24) + (r << 16) + (g << 8) + b;
}
return (a << 24) + (r << 16) + (g << 8) + b;
}
__global__ void drawColorDisp(uchar* disp, size_t disp_step, uchar* out_image, size_t out_step, int width, int height, int ndisp)
{
const int x = (blockIdx.x * blockDim.x + threadIdx.x) << 2;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if(x < width && y < height)
if(x < width && y < height)
{
uchar4 d4 = *(uchar4*)(disp + y * disp_step + x);
@@ -259,7 +259,7 @@ namespace cv { namespace gpu { namespace device
res.y = cvtPixel(d4.y, ndisp);
res.z = cvtPixel(d4.z, ndisp);
res.w = cvtPixel(d4.w, ndisp);
uint4* line = (uint4*)(out_image + y * out_step);
line[x >> 2] = res;
}
@@ -270,12 +270,12 @@ namespace cv { namespace gpu { namespace device
const int x = (blockIdx.x * blockDim.x + threadIdx.x) << 1;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if(x < width && y < height)
if(x < width && y < height)
{
short2 d2 = *(short2*)(disp + y * disp_step + x);
uint2 res;
res.x = cvtPixel(d2.x, ndisp);
res.x = cvtPixel(d2.x, ndisp);
res.y = cvtPixel(d2.y, ndisp);
uint2* line = (uint2*)(out_image + y * out_step);
@@ -290,12 +290,12 @@ namespace cv { namespace gpu { namespace device
dim3 grid(1, 1, 1);
grid.x = divUp(src.cols, threads.x << 2);
grid.y = divUp(src.rows, threads.y);
drawColorDisp<<<grid, threads, 0, stream>>>(src.data, src.step, dst.data, dst.step, src.cols, src.rows, ndisp);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaDeviceSynchronize() );
}
void drawColorDisp_gpu(const DevMem2D_<short>& src, const DevMem2Db& dst, int ndisp, const cudaStream_t& stream)
@@ -304,10 +304,10 @@ namespace cv { namespace gpu { namespace device
dim3 grid(1, 1, 1);
grid.x = divUp(src.cols, threads.x << 1);
grid.y = divUp(src.rows, threads.y);
drawColorDisp<<<grid, threads, 0, stream>>>(src.data, src.step / sizeof(short), dst.data, dst.step, src.cols, src.rows, ndisp);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
@@ -318,7 +318,7 @@ namespace cv { namespace gpu { namespace device
template <typename T>
__global__ void reprojectImageTo3D(const T* disp, size_t disp_step, float* xyzw, size_t xyzw_step, int rows, int cols)
{
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -328,7 +328,7 @@ namespace cv { namespace gpu { namespace device
float qx = cq[1] * y + cq[3], qy = cq[5] * y + cq[7];
float qz = cq[9] * y + cq[11], qw = cq[13] * y + cq[15];
qx += x * cq[0];
qx += x * cq[0];
qy += x * cq[4];
qz += x * cq[8];
qw += x * cq[12];
@@ -457,7 +457,7 @@ namespace cv { namespace gpu { namespace device
bindTexture(&harrisDxTex, Dx);
bindTexture(&harrisDyTex, Dy);
switch (border_type)
switch (border_type)
{
case BORDER_REFLECT101_GPU:
cornerHarris_kernel<<<grid, block, 0, stream>>>(block_size, k, dst, BrdRowReflect101<void>(Dx.cols), BrdColReflect101<void>(Dx.rows));
@@ -565,7 +565,7 @@ namespace cv { namespace gpu { namespace device
{
dim3 block(32, 8);
dim3 grid(divUp(Dx.cols, block.x), divUp(Dx.rows, block.y));
bindTexture(&minEigenValDxTex, Dx);
bindTexture(&minEigenValDyTex, Dy);
@@ -630,10 +630,10 @@ namespace cv { namespace gpu { namespace device
__global__ void mulSpectrumsKernel(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, DevMem2D_<cufftComplex> c)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < c.cols && y < c.rows)
if (x < c.cols && y < c.rows)
{
c.ptr(y)[x] = cuCmulf(a.ptr(y)[x], b.ptr(y)[x]);
}
@@ -658,10 +658,10 @@ namespace cv { namespace gpu { namespace device
__global__ void mulSpectrumsKernel_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, DevMem2D_<cufftComplex> c)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < c.cols && y < c.rows)
if (x < c.cols && y < c.rows)
{
c.ptr(y)[x] = cuCmulf(a.ptr(y)[x], cuConjf(b.ptr(y)[x]));
}
@@ -689,7 +689,7 @@ namespace cv { namespace gpu { namespace device
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < c.cols && y < c.rows)
if (x < c.cols && y < c.rows)
{
cufftComplex v = cuCmulf(a.ptr(y)[x], b.ptr(y)[x]);
c.ptr(y)[x] = make_cuFloatComplex(cuCrealf(v) * scale, cuCimagf(v) * scale);
@@ -718,7 +718,7 @@ namespace cv { namespace gpu { namespace device
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x < c.cols && y < c.rows)
if (x < c.cols && y < c.rows)
{
cufftComplex v = cuCmulf(a.ptr(y)[x], cuConjf(b.ptr(y)[x]));
c.ptr(y)[x] = make_cuFloatComplex(cuCrealf(v) * scale, cuCimagf(v) * scale);
@@ -736,7 +736,7 @@ namespace cv { namespace gpu { namespace device
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
//////////////////////////////////////////////////////////////////////////
// buildWarpMaps
@@ -842,7 +842,7 @@ namespace cv { namespace gpu { namespace device
void buildWarpPlaneMaps(int tl_u, int tl_v, DevMem2Df map_x, DevMem2Df map_y,
const float k_rinv[9], const float r_kinv[9], const float t[3],
const float k_rinv[9], const float r_kinv[9], const float t[3],
float scale, cudaStream_t stream)
{
cudaSafeCall(cudaMemcpyToSymbol(build_warp_maps::ck_rinv, k_rinv, 9*sizeof(float)));
@@ -911,27 +911,28 @@ namespace cv { namespace gpu { namespace device
__constant__ float c_filter2DKernel[FILTER2D_MAX_KERNEL_SIZE * FILTER2D_MAX_KERNEL_SIZE];
texture<float, cudaTextureType2D, cudaReadModeElementType> filter2DTex(0, cudaFilterModePoint, cudaAddressModeBorder);
texture<float, cudaTextureType2D, cudaReadModeElementType> filter2DTex(0, cudaFilterModePoint, cudaAddressModeClamp);
__global__ void filter2D(int ofsX, int ofsY, DevMem2Df dst, const int kWidth, const int kHeight, const int anchorX, const int anchorY)
__global__ void filter2D(int ofsX, int ofsY, PtrStepf dst, const int kWidth, const int kHeight, const int anchorX, const int anchorY, const BrdReflect101<float> brd)
{
const int x = blockIdx.x * blockDim.x + threadIdx.x;
const int y = blockIdx.y * blockDim.y + threadIdx.y;
if (x >= dst.cols || y >= dst.rows)
if (x > brd.last_col || y > brd.last_row)
return;
float res = 0;
const int baseX = ofsX + x - anchorX;
const int baseY = ofsY + y - anchorY;
int kInd = 0;
for (int i = 0; i < kHeight; ++i)
{
for (int j = 0; j < kWidth; ++j)
res += tex2D(filter2DTex, baseX + j, baseY + i) * c_filter2DKernel[kInd++];
{
const int srcX = ofsX + brd.idx_col(x - anchorX + j);
const int srcY = ofsY + brd.idx_row(y - anchorY + i);
res += tex2D(filter2DTex, srcX, srcY) * c_filter2DKernel[kInd++];
}
}
dst.ptr(y)[x] = res;
@@ -946,7 +947,9 @@ namespace cv { namespace gpu { namespace device
bindTexture(&filter2DTex, src);
filter2D<<<grid, block, 0, stream>>>(ofsX, ofsY, dst, kWidth, kHeight, anchorX, anchorY);
BrdReflect101<float> brd(dst.rows, dst.cols);
filter2D<<<grid, block, 0, stream>>>(ofsX, ofsY, dst, kWidth, kHeight, anchorX, anchorY, brd);
cudaSafeCall(cudaGetLastError());
if (stream == 0)

132
modules/gpu/src/filtering.cpp
View File

@@ -119,7 +119,7 @@ namespace
{
int scale = nDivisor && (kernel.depth() == CV_32F || kernel.depth() == CV_64F) ? 256 : 1;
if (nDivisor) *nDivisor = scale;
Mat temp(kernel.size(), type);
kernel.convertTo(temp, type, scale);
Mat cont_krnl = temp.reshape(1, 1);
@@ -134,7 +134,7 @@ namespace
}
gpu_krnl.upload(cont_krnl);
}
}
}
////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -144,7 +144,7 @@ namespace
{
struct Filter2DEngine_GPU : public FilterEngine_GPU
{
Filter2DEngine_GPU(const Ptr<BaseFilter_GPU>& filter2D_, int srcType_, int dstType_) :
Filter2DEngine_GPU(const Ptr<BaseFilter_GPU>& filter2D_, int srcType_, int dstType_) :
filter2D(filter2D_), srcType(srcType_), dstType(dstType_)
{}
@@ -189,9 +189,9 @@ namespace
{
struct SeparableFilterEngine_GPU : public FilterEngine_GPU
{
SeparableFilterEngine_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter_, const Ptr<BaseColumnFilter_GPU>& columnFilter_,
SeparableFilterEngine_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter_, const Ptr<BaseColumnFilter_GPU>& columnFilter_,
int srcType_, int bufType_, int dstType_) :
rowFilter(rowFilter_), columnFilter(columnFilter_),
rowFilter(rowFilter_), columnFilter(columnFilter_),
srcType(srcType_), bufType(bufType_), dstType(dstType_)
{
ksize = Size(rowFilter->ksize, columnFilter->ksize);
@@ -199,11 +199,11 @@ namespace
pbuf = &buf;
}
SeparableFilterEngine_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter_, const Ptr<BaseColumnFilter_GPU>& columnFilter_,
SeparableFilterEngine_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter_, const Ptr<BaseColumnFilter_GPU>& columnFilter_,
int srcType_, int bufType_, int dstType_,
GpuMat& buf_) :
rowFilter(rowFilter_), columnFilter(columnFilter_),
rowFilter(rowFilter_), columnFilter(columnFilter_),
srcType(srcType_), bufType(bufType_), dstType(dstType_)
{
ksize = Size(rowFilter->ksize, columnFilter->ksize);
@@ -235,7 +235,7 @@ namespace
GpuMat srcROI = src(roi);
GpuMat dstROI = dst(roi);
GpuMat bufROI = (*pbuf)(roi);
(*rowFilter)(srcROI, bufROI, stream);
(*columnFilter)(bufROI, dstROI, stream);
}
@@ -253,13 +253,13 @@ namespace
};
}
Ptr<FilterEngine_GPU> cv::gpu::createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter,
Ptr<FilterEngine_GPU> cv::gpu::createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter,
const Ptr<BaseColumnFilter_GPU>& columnFilter, int srcType, int bufType, int dstType)
{
return Ptr<FilterEngine_GPU>(new SeparableFilterEngine_GPU(rowFilter, columnFilter, srcType, bufType, dstType));
}
Ptr<FilterEngine_GPU> cv::gpu::createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter,
Ptr<FilterEngine_GPU> cv::gpu::createSeparableFilter_GPU(const Ptr<BaseRowFilter_GPU>& rowFilter,
const Ptr<BaseColumnFilter_GPU>& columnFilter, int srcType, int bufType, int dstType, GpuMat& buf)
{
return Ptr<FilterEngine_GPU>(new SeparableFilterEngine_GPU(rowFilter, columnFilter, srcType, bufType, dstType, buf));
@@ -284,7 +284,7 @@ namespace
NppStreamHandler h(stream);
nppSafeCall( nppiSumWindowRow_8u32f_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
nppSafeCall( nppiSumWindowRow_8u32f_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz, ksize, anchor) );
if (stream == 0)
@@ -318,7 +318,7 @@ namespace
NppStreamHandler h(stream);
nppSafeCall( nppiSumWindowColumn_8u32f_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
nppSafeCall( nppiSumWindowColumn_8u32f_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz, ksize, anchor) );
if (stream == 0)
@@ -341,7 +341,7 @@ Ptr<BaseColumnFilter_GPU> cv::gpu::getColumnSumFilter_GPU(int sumType, int dstTy
namespace
{
typedef NppStatus (*nppFilterBox_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oSizeROI,
typedef NppStatus (*nppFilterBox_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oSizeROI,
NppiSize oMaskSize, NppiPoint oAnchor);
struct NPPBoxFilter : public BaseFilter_GPU
@@ -363,8 +363,8 @@ namespace
cudaStream_t stream = StreamAccessor::getStream(s);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step),
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step),
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, oKernelSize, oAnchor) );
if (stream == 0)
@@ -379,7 +379,7 @@ Ptr<BaseFilter_GPU> cv::gpu::getBoxFilter_GPU(int srcType, int dstType, const Si
{
static const nppFilterBox_t nppFilterBox_callers[] = {0, nppiFilterBox_8u_C1R, 0, 0, nppiFilterBox_8u_C4R};
CV_Assert((srcType == CV_8UC1 || srcType == CV_8UC4) && dstType == srcType);
CV_Assert((srcType == CV_8UC1 || srcType == CV_8UC4) && dstType == srcType);
normalizeAnchor(anchor, ksize);
@@ -413,7 +413,7 @@ namespace
struct NPPMorphFilter : public BaseFilter_GPU
{
NPPMorphFilter(const Size& ksize_, const Point& anchor_, const GpuMat& kernel_, nppMorfFilter_t func_) :
NPPMorphFilter(const Size& ksize_, const Point& anchor_, const GpuMat& kernel_, nppMorfFilter_t func_) :
BaseFilter_GPU(ksize_, anchor_), kernel(kernel_), func(func_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& s = Stream::Null())
@@ -432,7 +432,7 @@ namespace
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step),
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step),
dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, kernel.ptr<Npp8u>(), oKernelSize, oAnchor) );
if (stream == 0)
@@ -446,19 +446,19 @@ namespace
Ptr<BaseFilter_GPU> cv::gpu::getMorphologyFilter_GPU(int op, int type, const Mat& kernel, const Size& ksize, Point anchor)
{
static const nppMorfFilter_t nppMorfFilter_callers[2][5] =
static const nppMorfFilter_t nppMorfFilter_callers[2][5] =
{
{0, nppiErode_8u_C1R, 0, 0, nppiErode_8u_C4R },
{0, nppiDilate_8u_C1R, 0, 0, nppiDilate_8u_C4R }
};
CV_Assert(op == MORPH_ERODE || op == MORPH_DILATE);
CV_Assert(type == CV_8UC1 || type == CV_8UC4);
CV_Assert(op == MORPH_ERODE || op == MORPH_DILATE);
CV_Assert(type == CV_8UC1 || type == CV_8UC4);
GpuMat gpu_krnl;
normalizeKernel(kernel, gpu_krnl);
normalizeAnchor(anchor, ksize);
return Ptr<BaseFilter_GPU>(new NPPMorphFilter(ksize, anchor, gpu_krnl, nppMorfFilter_callers[op][CV_MAT_CN(type)]));
}
@@ -466,13 +466,13 @@ namespace
{
struct MorphologyFilterEngine_GPU : public FilterEngine_GPU
{
MorphologyFilterEngine_GPU(const Ptr<BaseFilter_GPU>& filter2D_, int type_, int iters_) :
MorphologyFilterEngine_GPU(const Ptr<BaseFilter_GPU>& filter2D_, int type_, int iters_) :
filter2D(filter2D_), type(type_), iters(iters_)
{
pbuf = &buf;
}
MorphologyFilterEngine_GPU(const Ptr<BaseFilter_GPU>& filter2D_, int type_, int iters_, GpuMat& buf_) :
MorphologyFilterEngine_GPU(const Ptr<BaseFilter_GPU>& filter2D_, int type_, int iters_, GpuMat& buf_) :
filter2D(filter2D_), type(type_), iters(iters_)
{
pbuf = &buf_;
@@ -576,7 +576,7 @@ namespace
else if (iterations > 1 && countNonZero(_kernel) == _kernel.rows * _kernel.cols)
{
anchor = Point(anchor.x * iterations, anchor.y * iterations);
kernel = getStructuringElement(MORPH_RECT, Size(ksize.width + iterations * (ksize.width - 1),
kernel = getStructuringElement(MORPH_RECT, Size(ksize.width + iterations * (ksize.width - 1),
ksize.height + iterations * (ksize.height - 1)), anchor);
iterations = 1;
}
@@ -659,7 +659,7 @@ void cv::gpu::morphologyEx(const GpuMat& src, GpuMat& dst, int op, const Mat& ke
////////////////////////////////////////////////////////////////////////////////////////////////////
// Linear Filter
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace imgproc
{
@@ -669,12 +669,12 @@ namespace cv { namespace gpu { namespace device
namespace
{
typedef NppStatus (*nppFilter2D_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oSizeROI,
typedef NppStatus (*nppFilter2D_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oSizeROI,
const Npp32s * pKernel, NppiSize oKernelSize, NppiPoint oAnchor, Npp32s nDivisor);
struct NPPLinearFilter : public BaseFilter_GPU
{
NPPLinearFilter(const Size& ksize_, const Point& anchor_, const GpuMat& kernel_, Npp32s nDivisor_, nppFilter2D_t func_) :
NPPLinearFilter(const Size& ksize_, const Point& anchor_, const GpuMat& kernel_, Npp32s nDivisor_, nppFilter2D_t func_) :
BaseFilter_GPU(ksize_, anchor_), kernel(kernel_), nDivisor(nDivisor_), func(func_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& s = Stream::Null())
@@ -692,8 +692,8 @@ namespace
cudaStream_t stream = StreamAccessor::getStream(s);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz,
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz,
kernel.ptr<Npp32s>(), oKernelSize, oAnchor, nDivisor) );
if (stream == 0)
@@ -707,9 +707,9 @@ namespace
struct GpuLinearFilter : public BaseFilter_GPU
{
GpuLinearFilter(Size ksize_, Point anchor_, const GpuMat& kernel_) :
GpuLinearFilter(Size ksize_, Point anchor_, const GpuMat& kernel_) :
BaseFilter_GPU(ksize_, anchor_), kernel(kernel_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null())
{
using namespace cv::gpu::device::imgproc;
@@ -745,7 +745,7 @@ Ptr<BaseFilter_GPU> cv::gpu::getLinearFilter_GPU(int srcType, int dstType, const
else
{
static const nppFilter2D_t cppFilter2D_callers[] = {0, nppiFilter_8u_C1R, 0, 0, nppiFilter_8u_C4R};
GpuMat gpu_krnl;
int nDivisor;
normalizeKernel(kernel, gpu_krnl, CV_32S, &nDivisor, true);
@@ -753,8 +753,8 @@ Ptr<BaseFilter_GPU> cv::gpu::getLinearFilter_GPU(int srcType, int dstType, const
normalizeAnchor(anchor, ksize);
return Ptr<BaseFilter_GPU>(new NPPLinearFilter(ksize, anchor, gpu_krnl, nDivisor, cppFilter2D_callers[CV_MAT_CN(srcType)]));
}
}
}
}
Ptr<FilterEngine_GPU> cv::gpu::createLinearFilter_GPU(int srcType, int dstType, const Mat& kernel, const Point& anchor)
{
@@ -780,7 +780,7 @@ void cv::gpu::filter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& ke
////////////////////////////////////////////////////////////////////////////////////////////////////
// Separable Linear Filter
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace row_filter
{
@@ -797,14 +797,14 @@ namespace cv { namespace gpu { namespace device
namespace
{
typedef NppStatus (*nppFilter1D_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oROI,
typedef NppStatus (*nppFilter1D_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oROI,
const Npp32s * pKernel, Npp32s nMaskSize, Npp32s nAnchor, Npp32s nDivisor);
typedef void (*gpuFilter1D_t)(DevMem2Db src, DevMem2Db dst, const float kernel[], int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
struct NppLinearRowFilter : public BaseRowFilter_GPU
{
NppLinearRowFilter(int ksize_, int anchor_, const GpuMat& kernel_, Npp32s nDivisor_, nppFilter1D_t func_) :
NppLinearRowFilter(int ksize_, int anchor_, const GpuMat& kernel_, Npp32s nDivisor_, nppFilter1D_t func_) :
BaseRowFilter_GPU(ksize_, anchor_), kernel(kernel_), nDivisor(nDivisor_), func(func_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& s = Stream::Null())
@@ -817,7 +817,7 @@ namespace
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz,
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz,
kernel.ptr<Npp32s>(), ksize, anchor, nDivisor) );
if (stream == 0)
@@ -831,7 +831,7 @@ namespace
struct GpuLinearRowFilter : public BaseRowFilter_GPU
{
GpuLinearRowFilter(int ksize_, int anchor_, const Mat& kernel_, gpuFilter1D_t func_, int brd_type_) :
GpuLinearRowFilter(int ksize_, int anchor_, const Mat& kernel_, gpuFilter1D_t func_, int brd_type_) :
BaseRowFilter_GPU(ksize_, anchor_), kernel(kernel_), func(func_), brd_type(brd_type_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& s = Stream::Null())
@@ -852,7 +852,7 @@ Ptr<BaseRowFilter_GPU> cv::gpu::getLinearRowFilter_GPU(int srcType, int bufType,
using namespace ::cv::gpu::device::row_filter;
static const nppFilter1D_t nppFilter1D_callers[] = {0, nppiFilterRow_8u_C1R, 0, 0, nppiFilterRow_8u_C4R};
if ((bufType == srcType) && (srcType == CV_8UC1 || srcType == CV_8UC4))
{
CV_Assert(borderType == BORDER_CONSTANT);
@@ -867,7 +867,7 @@ Ptr<BaseRowFilter_GPU> cv::gpu::getLinearRowFilter_GPU(int srcType, int bufType,
return Ptr<BaseRowFilter_GPU>(new NppLinearRowFilter(ksize, anchor, gpu_row_krnl, nDivisor,
nppFilter1D_callers[CV_MAT_CN(srcType)]));
}
CV_Assert(borderType == BORDER_REFLECT101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderType, gpuBorderType));
@@ -914,7 +914,7 @@ namespace
{
struct NppLinearColumnFilter : public BaseColumnFilter_GPU
{
NppLinearColumnFilter(int ksize_, int anchor_, const GpuMat& kernel_, Npp32s nDivisor_, nppFilter1D_t func_) :
NppLinearColumnFilter(int ksize_, int anchor_, const GpuMat& kernel_, Npp32s nDivisor_, nppFilter1D_t func_) :
BaseColumnFilter_GPU(ksize_, anchor_), kernel(kernel_), nDivisor(nDivisor_), func(func_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& s = Stream::Null())
@@ -927,7 +927,7 @@ namespace
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz,
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz,
kernel.ptr<Npp32s>(), ksize, anchor, nDivisor) );
if (stream == 0)
@@ -941,7 +941,7 @@ namespace
struct GpuLinearColumnFilter : public BaseColumnFilter_GPU
{
GpuLinearColumnFilter(int ksize_, int anchor_, const Mat& kernel_, gpuFilter1D_t func_, int brd_type_) :
GpuLinearColumnFilter(int ksize_, int anchor_, const Mat& kernel_, gpuFilter1D_t func_, int brd_type_) :
BaseColumnFilter_GPU(ksize_, anchor_), kernel(kernel_), func(func_), brd_type(brd_type_) {}
virtual void operator()(const GpuMat& src, GpuMat& dst, Stream& s = Stream::Null())
@@ -963,7 +963,7 @@ Ptr<BaseColumnFilter_GPU> cv::gpu::getLinearColumnFilter_GPU(int bufType, int ds
using namespace ::cv::gpu::device::column_filter;
static const nppFilter1D_t nppFilter1D_callers[] = {0, nppiFilterColumn_8u_C1R, 0, 0, nppiFilterColumn_8u_C4R};
if ((bufType == dstType) && (bufType == CV_8UC1 || bufType == CV_8UC4))
{
CV_Assert(borderType == BORDER_CONSTANT);
@@ -975,14 +975,14 @@ Ptr<BaseColumnFilter_GPU> cv::gpu::getLinearColumnFilter_GPU(int bufType, int ds
int ksize = gpu_col_krnl.cols;
normalizeAnchor(anchor, ksize);
return Ptr<BaseColumnFilter_GPU>(new NppLinearColumnFilter(ksize, anchor, gpu_col_krnl, nDivisor,
return Ptr<BaseColumnFilter_GPU>(new NppLinearColumnFilter(ksize, anchor, gpu_col_krnl, nDivisor,
nppFilter1D_callers[CV_MAT_CN(bufType)]));
}
CV_Assert(borderType == BORDER_REFLECT101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP);
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderType, gpuBorderType));
CV_Assert(dstType == CV_8UC1 || dstType == CV_8UC4 || dstType == CV_16SC3 || dstType == CV_32SC1 || dstType == CV_32FC1);
CV_Assert(CV_MAT_DEPTH(bufType) == CV_32F && CV_MAT_CN(dstType) == CV_MAT_CN(bufType));
@@ -1021,7 +1021,7 @@ Ptr<BaseColumnFilter_GPU> cv::gpu::getLinearColumnFilter_GPU(int bufType, int ds
return Ptr<BaseColumnFilter_GPU>(new GpuLinearColumnFilter(ksize, anchor, cont_krnl, func, gpuBorderType));
}
Ptr<FilterEngine_GPU> cv::gpu::createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat& rowKernel, const Mat& columnKernel,
Ptr<FilterEngine_GPU> cv::gpu::createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat& rowKernel, const Mat& columnKernel,
const Point& anchor, int rowBorderType, int columnBorderType)
{
if (columnBorderType < 0)
@@ -1037,7 +1037,7 @@ Ptr<FilterEngine_GPU> cv::gpu::createSeparableLinearFilter_GPU(int srcType, int
return createSeparableFilter_GPU(rowFilter, columnFilter, srcType, bufType, dstType);
}
Ptr<FilterEngine_GPU> cv::gpu::createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat& rowKernel, const Mat& columnKernel, GpuMat& buf,
Ptr<FilterEngine_GPU> cv::gpu::createSeparableLinearFilter_GPU(int srcType, int dstType, const Mat& rowKernel, const Mat& columnKernel, GpuMat& buf,
const Point& anchor, int rowBorderType, int columnBorderType)
{
if (columnBorderType < 0)
@@ -1053,7 +1053,7 @@ Ptr<FilterEngine_GPU> cv::gpu::createSeparableLinearFilter_GPU(int srcType, int
return createSeparableFilter_GPU(rowFilter, columnFilter, srcType, bufType, dstType, buf);
}
void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernelX, const Mat& kernelY,
void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernelX, const Mat& kernelY,
Point anchor, int rowBorderType, int columnBorderType)
{
if( ddepth < 0 )
@@ -1065,7 +1065,7 @@ void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat&
f->apply(src, dst, Rect(0, 0, src.cols, src.rows));
}
void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernelX, const Mat& kernelY, GpuMat& buf,
void cv::gpu::sepFilter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& kernelX, const Mat& kernelY, GpuMat& buf,
Point anchor, int rowBorderType, int columnBorderType,
Stream& stream)
{
@@ -1115,7 +1115,7 @@ void cv::gpu::Sobel(const GpuMat& src, GpuMat& dst, int ddepth, int dx, int dy,
else
ky *= scale;
}
sepFilter2D(src, dst, ddepth, kx, ky, buf, Point(-1,-1), rowBorderType, columnBorderType, stream);
}
@@ -1155,7 +1155,7 @@ void cv::gpu::Laplacian(const GpuMat& src, GpuMat& dst, int ddepth, int ksize, d
Mat kernel(3, 3, CV_32S, (void*)K[ksize == 3]);
if (scale != 1)
kernel *= scale;
filter2D(src, dst, ddepth, kernel, Point(-1,-1), stream);
}
@@ -1163,7 +1163,7 @@ void cv::gpu::Laplacian(const GpuMat& src, GpuMat& dst, int ddepth, int ksize, d
// Gaussian Filter
Ptr<FilterEngine_GPU> cv::gpu::createGaussianFilter_GPU(int type, Size ksize, double sigma1, double sigma2, int rowBorderType, int columnBorderType)
{
{
int depth = CV_MAT_DEPTH(type);
if (sigma2 <= 0)
@@ -1191,7 +1191,7 @@ Ptr<FilterEngine_GPU> cv::gpu::createGaussianFilter_GPU(int type, Size ksize, do
}
Ptr<FilterEngine_GPU> cv::gpu::createGaussianFilter_GPU(int type, Size ksize, GpuMat& buf, double sigma1, double sigma2, int rowBorderType, int columnBorderType)
{
{
int depth = CV_MAT_DEPTH(type);
if (sigma2 <= 0)
@@ -1227,7 +1227,7 @@ void cv::gpu::GaussianBlur(const GpuMat& src, GpuMat& dst, Size ksize, double si
}
dst.create(src.size(), src.type());
Ptr<FilterEngine_GPU> f = createGaussianFilter_GPU(src.type(), ksize, sigma1, sigma2, rowBorderType, columnBorderType);
f->apply(src, dst, Rect(0, 0, src.cols, src.rows));
}
@@ -1241,7 +1241,7 @@ void cv::gpu::GaussianBlur(const GpuMat& src, GpuMat& dst, Size ksize, GpuMat& b
}
dst.create(src.size(), src.type());
Ptr<FilterEngine_GPU> f = createGaussianFilter_GPU(src.type(), ksize, buf, sigma1, sigma2, rowBorderType, columnBorderType);
f->apply(src, dst, Rect(0, 0, src.cols, src.rows), stream);
}
@@ -1251,7 +1251,7 @@ void cv::gpu::GaussianBlur(const GpuMat& src, GpuMat& dst, Size ksize, GpuMat& b
namespace
{
typedef NppStatus (*nppFilterRank_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oSizeROI,
typedef NppStatus (*nppFilterRank_t)(const Npp8u * pSrc, Npp32s nSrcStep, Npp8u * pDst, Npp32s nDstStep, NppiSize oSizeROI,
NppiSize oMaskSize, NppiPoint oAnchor);
struct NPPRankFilter : public BaseFilter_GPU
@@ -1273,7 +1273,7 @@ namespace
cudaStream_t stream = StreamAccessor::getStream(s);
NppStreamHandler h(stream);
nppSafeCall( func(src.ptr<Npp8u>(), static_cast<int>(src.step), dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, oKernelSize, oAnchor) );
if (stream == 0)
@@ -1288,7 +1288,7 @@ Ptr<BaseFilter_GPU> cv::gpu::getMaxFilter_GPU(int srcType, int dstType, const Si
{
static const nppFilterRank_t nppFilterRank_callers[] = {0, nppiFilterMax_8u_C1R, 0, 0, nppiFilterMax_8u_C4R};
CV_Assert((srcType == CV_8UC1 || srcType == CV_8UC4) && dstType == srcType);
CV_Assert((srcType == CV_8UC1 || srcType == CV_8UC4) && dstType == srcType);
normalizeAnchor(anchor, ksize);
@@ -1299,7 +1299,7 @@ Ptr<BaseFilter_GPU> cv::gpu::getMinFilter_GPU(int srcType, int dstType, const Si
{
static const nppFilterRank_t nppFilterRank_callers[] = {0, nppiFilterMin_8u_C1R, 0, 0, nppiFilterMin_8u_C4R};
CV_Assert((srcType == CV_8UC1 || srcType == CV_8UC4) && dstType == srcType);
CV_Assert((srcType == CV_8UC1 || srcType == CV_8UC4) && dstType == srcType);
normalizeAnchor(anchor, ksize);

38
modules/gpu/src/surf.cpp
View File

@@ -63,7 +63,7 @@ void cv::gpu::SURF_GPU::releaseMemory() { throw_nogpu(); }
#else /* !defined (HAVE_CUDA) */
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace surf
{
@@ -79,13 +79,13 @@ namespace cv { namespace gpu { namespace device
void icvFindMaximaInLayer_gpu(const PtrStepf& det, const PtrStepf& trace, int4* maxPosBuffer, unsigned int* maxCounter,
int img_rows, int img_cols, int octave, bool use_mask, int nLayers);
void icvInterpolateKeypoint_gpu(const PtrStepf& det, const int4* maxPosBuffer, unsigned int maxCounter,
float* featureX, float* featureY, int* featureLaplacian, float* featureSize, float* featureHessian,
void icvInterpolateKeypoint_gpu(const PtrStepf& det, const int4* maxPosBuffer, unsigned int maxCounter,
float* featureX, float* featureY, int* featureLaplacian, float* featureSize, float* featureHessian,
unsigned int* featureCounter);
void icvCalcOrientation_gpu(const float* featureX, const float* featureY, const float* featureSize, float* featureDir, int nFeatures);
void compute_descriptors_gpu(const DevMem2Df& descriptors,
void compute_descriptors_gpu(const DevMem2Df& descriptors,
const float* featureX, const float* featureY, const float* featureSize, const float* featureDir, int nFeatures);
}
}}}
@@ -108,7 +108,7 @@ namespace
return (HAAR_SIZE0 + HAAR_SIZE_INC * layer) << octave;
}
class SURF_GPU_Invoker
{
public:
@@ -121,11 +121,11 @@ namespace
CV_Assert(mask.empty() || (mask.size() == img.size() && mask.type() == CV_8UC1));
CV_Assert(surf_.nOctaves > 0 && surf_.nOctaveLayers > 0);
CV_Assert(TargetArchs::builtWith(GLOBAL_ATOMICS) && DeviceInfo().supports(GLOBAL_ATOMICS));
const int min_size = calcSize(surf_.nOctaves - 1, 0);
CV_Assert(img_rows - min_size >= 0);
CV_Assert(img_cols - min_size >= 0);
const int layer_rows = img_rows >> (surf_.nOctaves - 1);
const int layer_cols = img_cols >> (surf_.nOctaves - 1);
const int min_margin = ((calcSize((surf_.nOctaves - 1), 2) >> 1) >> (surf_.nOctaves - 1)) + 1;
@@ -159,7 +159,7 @@ namespace
{
ensureSizeIsEnough(img_rows * (surf_.nOctaveLayers + 2), img_cols, CV_32FC1, surf_.det);
ensureSizeIsEnough(img_rows * (surf_.nOctaveLayers + 2), img_cols, CV_32FC1, surf_.trace);
ensureSizeIsEnough(1, maxCandidates, CV_32SC4, surf_.maxPosBuffer);
ensureSizeIsEnough(SURF_GPU::SF_FEATURE_STRIDE, maxFeatures, CV_32FC1, keypoints);
keypoints.setTo(Scalar::all(0));
@@ -182,7 +182,7 @@ namespace
if (maxCounter > 0)
{
icvInterpolateKeypoint_gpu(surf_.det, surf_.maxPosBuffer.ptr<int4>(), maxCounter,
icvInterpolateKeypoint_gpu(surf_.det, surf_.maxPosBuffer.ptr<int4>(), maxCounter,
keypoints.ptr<float>(SURF_GPU::SF_X), keypoints.ptr<float>(SURF_GPU::SF_Y),
keypoints.ptr<int>(SURF_GPU::SF_LAPLACIAN), keypoints.ptr<float>(SURF_GPU::SF_SIZE),
keypoints.ptr<float>(SURF_GPU::SF_HESSIAN), counters.ptr<unsigned int>());
@@ -238,7 +238,7 @@ namespace
cv::gpu::SURF_GPU::SURF_GPU()
{
hessianThreshold = 100;
extended = 1;
extended = true;
nOctaves = 4;
nOctaveLayers = 2;
keypointsRatio = 0.01f;
@@ -323,9 +323,9 @@ void cv::gpu::SURF_GPU::downloadKeypoints(const GpuMat& keypointsGPU, vector<Key
else
{
CV_Assert(keypointsGPU.type() == CV_32FC1 && keypointsGPU.rows == SF_FEATURE_STRIDE);
Mat keypointsCPU(keypointsGPU);
keypoints.resize(nFeatures);
float* kp_x = keypointsCPU.ptr<float>(SF_X);
@@ -373,13 +373,13 @@ void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, GpuMat
}
}
void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, GpuMat& keypoints, GpuMat& descriptors,
void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, GpuMat& keypoints, GpuMat& descriptors,
bool useProvidedKeypoints)
{
if (!img.empty())
{
SURF_GPU_Invoker surf(*this, img, mask);
if (!useProvidedKeypoints)
surf.detectKeypoints(keypoints);
else if (!upright)
@@ -400,20 +400,20 @@ void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, vector
downloadKeypoints(keypointsGPU, keypoints);
}
void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, vector<KeyPoint>& keypoints,
void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, vector<KeyPoint>& keypoints,
GpuMat& descriptors, bool useProvidedKeypoints)
{
GpuMat keypointsGPU;
if (useProvidedKeypoints)
uploadKeypoints(keypoints, keypointsGPU);
uploadKeypoints(keypoints, keypointsGPU);
(*this)(img, mask, keypointsGPU, descriptors, useProvidedKeypoints);
downloadKeypoints(keypointsGPU, keypoints);
}
void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, vector<KeyPoint>& keypoints,
void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, vector<KeyPoint>& keypoints,
vector<float>& descriptors, bool useProvidedKeypoints)
{
GpuMat descriptorsGPU;
@@ -423,9 +423,9 @@ void cv::gpu::SURF_GPU::operator()(const GpuMat& img, const GpuMat& mask, vector
downloadDescriptors(descriptorsGPU, descriptors);
}
void cv::gpu::SURF_GPU::releaseMemory()
void cv::gpu::SURF_GPU::releaseMemory()
{
sum.release();
sum.release();
mask1.release();
maskSum.release();
intBuffer.release();