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

cosmetic changes in gpu module, decreased matchTemplate test running time

Browse Source
This commit is contained in:
Alexey Spizhevoy
2011-01-12 09:30:08 +00:00
parent d091ae5746
commit 4dfbf99dd5
5 changed files with 50 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
modules/gpu/include/opencv2/gpu/gpu.hpp
View File

@@ -1182,8 +1182,8 @@ namespace cv
void setSVMDetector(const vector<float>& detector);
static vector<float> getDefaultPeopleDetector();
static vector<float> getPeopleDetector_48x96();
static vector<float> getPeopleDetector_64x128();
static vector<float> getPeopleDetector48x96();
static vector<float> getPeopleDetector64x128();
void detect(const GpuMat& img, vector<Point>& found_locations,
double hit_threshold=0, Size win_stride=Size(),

20
modules/gpu/src/cuda/match_template.cu
View File

@@ -560,7 +560,7 @@ __global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8U(
(image_sqsum.ptr(y + h)[x + w] - image_sqsum.ptr(y)[x + w]) -
(image_sqsum.ptr(y + h)[x] - image_sqsum.ptr(y)[x]));
result.ptr(y)[x] = (ccorr - image_sum_ * templ_sum_scale) *
rsqrtf(templ_sqsum_scale * max(1.f, image_sqsum_ - weight * image_sum_ * image_sum_));
rsqrtf(templ_sqsum_scale * max(1e-3f, image_sqsum_ - weight * image_sum_ * image_sum_));
}
}
@@ -610,8 +610,8 @@ __global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8UC2(
(image_sqsum_g.ptr(y + h)[x + w] - image_sqsum_g.ptr(y)[x + w]) -
(image_sqsum_g.ptr(y + h)[x] - image_sqsum_g.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
float rdenom = rsqrtf(templ_sqsum_scale * max(1.f, image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_));
float rdenom = rsqrtf(templ_sqsum_scale * max(1e-3f, image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_));
result.ptr(y)[x] = (ccorr - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g) * rdenom;
}
@@ -678,9 +678,9 @@ __global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8UC3(
(image_sqsum_b.ptr(y + h)[x + w] - image_sqsum_b.ptr(y)[x + w]) -
(image_sqsum_b.ptr(y + h)[x] - image_sqsum_b.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
float rdenom = rsqrtf(templ_sqsum_scale * max(1.f, image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_
+ image_sqsum_b_ - weight * image_sum_b_ * image_sum_b_));
float rdenom = rsqrtf(templ_sqsum_scale * max(1e-3f, image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_
+ image_sqsum_b_ - weight * image_sum_b_ * image_sum_b_));
result.ptr(y)[x] = (ccorr - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g
- image_sum_b_ * templ_sum_scale_b) * rdenom;
@@ -760,10 +760,10 @@ __global__ void matchTemplatePreparedKernel_CCOFF_NORMED_8UC4(
(image_sqsum_a.ptr(y + h)[x + w] - image_sqsum_a.ptr(y)[x + w]) -
(image_sqsum_a.ptr(y + h)[x] - image_sqsum_a.ptr(y)[x]));
float ccorr = result.ptr(y)[x];
float rdenom = rsqrtf(templ_sqsum_scale * max(1.f, image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_
+ image_sqsum_b_ - weight * image_sum_b_ * image_sum_b_
+ image_sqsum_a_ - weight * image_sum_a_ * image_sum_a_));
float rdenom = rsqrtf(templ_sqsum_scale * max(1e-3f, image_sqsum_r_ - weight * image_sum_r_ * image_sum_r_
+ image_sqsum_g_ - weight * image_sum_g_ * image_sum_g_
+ image_sqsum_b_ - weight * image_sum_b_ * image_sum_b_
+ image_sqsum_a_ - weight * image_sum_a_ * image_sum_a_));
result.ptr(y)[x] = (ccorr - image_sum_r_ * templ_sum_scale_r
- image_sum_g_ * templ_sum_scale_g
- image_sum_b_ * templ_sum_scale_b

10
modules/gpu/src/hog.cpp
View File

@@ -55,8 +55,8 @@ void cv::gpu::HOGDescriptor::detectMultiScale(const GpuMat&, vector<Rect>&, doub
void cv::gpu::HOGDescriptor::computeBlockHistograms(const GpuMat&) { throw_nogpu(); }
void cv::gpu::HOGDescriptor::getDescriptors(const GpuMat&, Size, GpuMat&, int) { throw_nogpu(); }
std::vector<float> cv::gpu::HOGDescriptor::getDefaultPeopleDetector() { throw_nogpu(); return std::vector<float>(); }
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_48x96() { throw_nogpu(); return std::vector<float>(); }
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_64x128() { throw_nogpu(); return std::vector<float>(); }
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector48x96() { throw_nogpu(); return std::vector<float>(); }
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector64x128() { throw_nogpu(); return std::vector<float>(); }
#else
@@ -352,11 +352,11 @@ cv::Size cv::gpu::HOGDescriptor::numPartsWithin(cv::Size size, cv::Size part_siz
std::vector<float> cv::gpu::HOGDescriptor::getDefaultPeopleDetector()
{
return getPeopleDetector_64x128();
return getPeopleDetector64x128();
}
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_48x96()
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector48x96()
{
static const float detector[] = {
0.294350f, -0.098796f, -0.129522f, 0.078753f, 0.387527f, 0.261529f,
@@ -696,7 +696,7 @@ std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_48x96()
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector_64x128()
std::vector<float> cv::gpu::HOGDescriptor::getPeopleDetector64x128()
{
static const float detector[] = {
0.05359386f, -0.14721455f, -0.05532170f, 0.05077307f,