potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 15:23:05 +04:00
Code Packages Releases Wiki Activity

replace offsets in surf to simple copy for better speed

Browse Source
This commit is contained in:
marina.kolpakova
2012-11-24 16:50:29 +04:00
parent be0c20b758
commit 68d04d28b6
3 changed files with 29 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/gpu/src/surf.cpp
+12 -6
View File
@@ -75,10 +75,10 @@ namespace cv { namespace gpu { namespace device
size_t bindMaskSumTex(PtrStepSz<unsigned int> maskSum);
void icvCalcLayerDetAndTrace_gpu(const PtrStepf& det, const PtrStepf& trace, int img_rows, int img_cols,
int octave, int nOctaveLayers, const size_t sumOffset);
int octave, int nOctaveLayer);
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, const size_t maskOffset);
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, int* featureOctave, float* featureSize, float* featureHessian,
@@ -146,14 +146,17 @@ namespace
loadGlobalConstants(maxCandidates, maxFeatures, img_rows, img_cols, surf_.nOctaveLayers, static_cast<float>(surf_.hessianThreshold));
bindImgTex(img);
integralBuffered(img, surf_.sum, surf_.intBuffer);
integralBuffered(img, tmpSum, surf_.intBuffer);
tmpSum.copyTo(surf_.sum);
sumOffset = bindSumTex(surf_.sum);
if (use_mask)
{
min(mask, 1.0, surf_.mask1);
integralBuffered(surf_.mask1, surf_.maskSum, surf_.intBuffer);
integralBuffered(surf_.mask1, tmpMaskSum, surf_.intBuffer);
tmpMaskSum.copyTo(surf_.maskSum);
maskOffset = bindMaskSumTex(surf_.maskSum);
}
}
@@ -174,10 +177,10 @@ namespace
loadOctaveConstants(octave, layer_rows, layer_cols);
icvCalcLayerDetAndTrace_gpu(surf_.det, surf_.trace, img_rows, img_cols, octave, surf_.nOctaveLayers, sumOffset);
icvCalcLayerDetAndTrace_gpu(surf_.det, surf_.trace, img_rows, img_cols, octave, surf_.nOctaveLayers);
icvFindMaximaInLayer_gpu(surf_.det, surf_.trace, surf_.maxPosBuffer.ptr<int4>(), counters.ptr<unsigned int>() + 1 + octave,
img_rows, img_cols, octave, use_mask, surf_.nOctaveLayers, maskOffset);
img_rows, img_cols, octave, use_mask, surf_.nOctaveLayers);
unsigned int maxCounter;
cudaSafeCall( cudaMemcpy(&maxCounter, counters.ptr<unsigned int>() + 1 + octave, sizeof(unsigned int), cudaMemcpyDeviceToHost) );
@@ -228,6 +231,9 @@ namespace
private:
SURF_GPU& surf_;
cv::gpu::GpuMat tmpSum;
cv::gpu::GpuMat tmpMaskSum;
int img_cols, img_rows;
bool use_mask;