gpu::add now supports 8UC4 and 32FC1

This commit is contained in:
Kirill Kornyakov 2010-09-10 15:04:54 +00:00
parent a2a3ec69cb
commit 8e38063965
3 changed files with 119 additions and 34 deletions

View File

@ -41,7 +41,6 @@
//M*/
#include "precomp.hpp"
#include "npp.h" //TODO: move to the precomp.hpp
using namespace cv;
using namespace cv::gpu;
@ -55,24 +54,39 @@ void cv::gpu::add(const GpuMat& src1, const GpuMat& src2, GpuMat& dst) { throw_n
void cv::gpu::add(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
dst.create( src1.size(), src1.type() );
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
dst.create( src1.size(), src1.type() );
CV_DbgAssert(src1.depth() == CV_8U || src1.depth() == CV_32F);
CV_DbgAssert(src1.channels() == 1 || src1.channels() == 4);
int nChannels = src1.channels();
CV_DbgAssert((src1.depth() == CV_8U && nChannels == 1 || nChannels == 4) ||
(src1.depth() == CV_32F && nChannels == 1));
NppiSize sz;
sz.width = src1.cols;
sz.width = src1.cols;
sz.height = src1.rows;
if (src1.depth() == CV_8U)
{
nppiAdd_8u_C1RSfs((const Npp8u*)src1.ptr<char>(), src1.step,
(const Npp8u*)src2.ptr<char>(), src2.step,
(Npp8u*)dst.ptr<char>(), dst.step, sz, 0);
if (nChannels == 1)
{
nppiAdd_8u_C1RSfs((const Npp8u*)src1.ptr<char>(), src1.step,
(const Npp8u*)src2.ptr<char>(), src2.step,
(Npp8u*)dst.ptr<char>(), dst.step, sz, 0);
}
else
{
nppiAdd_8u_C4RSfs((const Npp8u*)src1.ptr<char>(), src1.step,
(const Npp8u*)src2.ptr<char>(), src2.step,
(Npp8u*)dst.ptr<char>(), dst.step, sz, 0);
}
}
else //if (src1.depth() == CV_32F)
{
nppiAdd_32f_C1R((const Npp32f*)src1.ptr<float>(), src1.step,
(const Npp32f*)src2.ptr<float>(), src2.step,
(Npp32f*)dst.ptr<float>(), dst.step, sz);
}
//TODO: implement other depths
}
#endif /* !defined (HAVE_CUDA) */

View File

@ -62,6 +62,7 @@
#include "cuda_shared.hpp"
#include "cuda_runtime_api.h"
#include "opencv2/gpu/stream_accessor.hpp"
#include "npp.h"
#else /* defined(HAVE_CUDA) */

View File

@ -40,7 +40,8 @@
//M*/
#include "gputest.hpp"
#include "highgui.h"
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
using namespace cv;
using namespace std;
@ -54,6 +55,13 @@ public:
protected:
void run(int);
int test8UC1(const Mat& imgL, const Mat& imgR);
int test8UC4(const Mat& imgL, const Mat& imgR);
int test32FC1(const Mat& imgL, const Mat& imgR);
int test(const Mat& imgL, const Mat& imgR);
int CheckNorm(const Mat& m1, const Mat& m2);
};
CV_GpuNppImageAdditionTest::CV_GpuNppImageAdditionTest(): CvTest( "GPU-NppImageAddition", "add" )
@ -62,10 +70,72 @@ CV_GpuNppImageAdditionTest::CV_GpuNppImageAdditionTest(): CvTest( "GPU-NppImageA
CV_GpuNppImageAdditionTest::~CV_GpuNppImageAdditionTest() {}
int CV_GpuNppImageAdditionTest::test8UC1(const Mat& imgL, const Mat& imgR)
{
cv::Mat imgL_C1;
cv::Mat imgR_C1;
cvtColor(imgL, imgL_C1, CV_BGR2GRAY);
cvtColor(imgR, imgR_C1, CV_BGR2GRAY);
return test(imgL_C1, imgR_C1);
}
int CV_GpuNppImageAdditionTest::test8UC4(const Mat& imgL, const Mat& imgR)
{
cv::Mat imgL_C4;
cv::Mat imgR_C4;
cvtColor(imgL, imgL_C4, CV_BGR2BGRA);
cvtColor(imgR, imgR_C4, CV_BGR2BGRA);
return test(imgL_C4, imgR_C4);
}
int CV_GpuNppImageAdditionTest::test32FC1( const Mat& imgL, const Mat& imgR )
{
cv::Mat imgL_C1;
cv::Mat imgR_C1;
cvtColor(imgL, imgL_C1, CV_BGR2GRAY);
cvtColor(imgR, imgR_C1, CV_BGR2GRAY);
imgL_C1.convertTo(imgL_C1, CV_32F);
imgR_C1.convertTo(imgR_C1, CV_32F);
return test(imgL_C1, imgR_C1);
}
int CV_GpuNppImageAdditionTest::test( const Mat& imgL, const Mat& imgR )
{
cv::Mat cpuAdd;
cv::add(imgL, imgR, cpuAdd);
GpuMat gpuL(imgL);
GpuMat gpuR(imgR);
GpuMat gpuAdd;
cv::gpu::add(gpuL, gpuR, gpuAdd);
return CheckNorm(cpuAdd, gpuAdd);
}
int CV_GpuNppImageAdditionTest::CheckNorm(const Mat& m1, const Mat& m2)
{
double ret = norm(m1, m2);
if (ret < 1.0)
{
return CvTS::OK;
}
else
{
ts->printf(CvTS::LOG, "\nNorm: %f\n", ret);
return CvTS::FAIL_GENERIC;
}
}
void CV_GpuNppImageAdditionTest::run( int )
{
cv::Mat img_l = cv::imread(std::string(ts->get_data_path()) + "stereobm/aloe-L.png", 0);
cv::Mat img_r = cv::imread(std::string(ts->get_data_path()) + "stereobm/aloe-R.png", 0);
//load images
cv::Mat img_l = cv::imread(std::string(ts->get_data_path()) + "stereobm/aloe-L.png");
cv::Mat img_r = cv::imread(std::string(ts->get_data_path()) + "stereobm/aloe-R.png");
if (img_l.empty() || img_r.empty())
{
@ -73,29 +143,29 @@ void CV_GpuNppImageAdditionTest::run( int )
return;
}
cv::Mat cpuAdd;
cv::add(img_l, img_r, cpuAdd);
GpuMat gpuL(img_l);
GpuMat gpuR(img_r);
GpuMat gpuAdd;
cv::gpu::add(gpuL, gpuR, gpuAdd);
//namedWindow("gpu");
//imshow("gpu", gpuAdd);
//namedWindow("cpu");
//imshow("cpu", cpuAdd);
//waitKey(1000);
double ret = norm(cpuAdd, gpuAdd);
if (ret < 1.0)
ts->set_failed_test_info(CvTS::OK);
else
//run tests
int testResult = test8UC1(img_l, img_r);
if (testResult != CvTS::OK)
{
ts->printf(CvTS::CONSOLE, "\nNorm: %f\n", ret);
ts->set_failed_test_info(CvTS::FAIL_GENERIC);
ts->set_failed_test_info(testResult);
return;
}
testResult = test8UC4(img_l, img_r);
if (testResult != CvTS::OK)
{
ts->set_failed_test_info(testResult);
return;
}
testResult = test32FC1(img_l, img_r);
if (testResult != CvTS::OK)
{
ts->set_failed_test_info(testResult);
return;
}
ts->set_failed_test_info(CvTS::OK);
}
CV_GpuNppImageAdditionTest CV_GpuNppImageAddition_test;