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

refactored gpu perf tests and fixed sanity tests

Browse Source
This commit is contained in:
Vladislav Vinogradov 2013-02-26 13:49:35 +04:00
parent 0d12f451be
commit a138e12a26
18 changed files with 2366 additions and 2754 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

219
modules/gpu/perf/perf_calib3d.cpp
View File

@ -3,15 +3,14 @@
using namespace std;
using namespace testing;
namespace {
//////////////////////////////////////////////////////////////////////
// StereoBM
typedef std::tr1::tuple<string, string> pair_string;
DEF_PARAM_TEST_1(ImagePair, pair_string);
PERF_TEST_P(ImagePair, Calib3D_StereoBM, Values(pair_string("gpu/perf/aloe.png", "gpu/perf/aloeR.png")))
PERF_TEST_P(ImagePair, Calib3D_StereoBM,
Values(pair_string("gpu/perf/aloe.png", "gpu/perf/aloeR.png")))
{
declare.time(5.0);
@ -28,18 +27,13 @@ PERF_TEST_P(ImagePair, Calib3D_StereoBM, Values(pair_string("gpu/perf/aloe.png",
{
cv::gpu::StereoBM_GPU d_bm(preset, ndisp);
cv::gpu::GpuMat d_imgLeft(imgLeft);
cv::gpu::GpuMat d_imgRight(imgRight);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_imgLeft(imgLeft);
const cv::gpu::GpuMat d_imgRight(imgRight);
cv::gpu::GpuMat dst;
d_bm(d_imgLeft, d_imgRight, d_dst);
TEST_CYCLE() d_bm(d_imgLeft, d_imgRight, dst);
TEST_CYCLE()
{
d_bm(d_imgLeft, d_imgRight, d_dst);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
@ -47,12 +41,7 @@ PERF_TEST_P(ImagePair, Calib3D_StereoBM, Values(pair_string("gpu/perf/aloe.png",
cv::Mat dst;
bm(imgLeft, imgRight, dst);
TEST_CYCLE()
{
bm(imgLeft, imgRight, dst);
}
TEST_CYCLE() bm(imgLeft, imgRight, dst);
CPU_SANITY_CHECK(dst);
}
@ -61,7 +50,8 @@ PERF_TEST_P(ImagePair, Calib3D_StereoBM, Values(pair_string("gpu/perf/aloe.png",
//////////////////////////////////////////////////////////////////////
// StereoBeliefPropagation
PERF_TEST_P(ImagePair, Calib3D_StereoBeliefPropagation, Values(pair_string("gpu/stereobp/aloe-L.png", "gpu/stereobp/aloe-R.png")))
PERF_TEST_P(ImagePair, Calib3D_StereoBeliefPropagation,
Values(pair_string("gpu/stereobp/aloe-L.png", "gpu/stereobp/aloe-R.png")))
{
declare.time(10.0);
@ -77,29 +67,25 @@ PERF_TEST_P(ImagePair, Calib3D_StereoBeliefPropagation, Values(pair_string("gpu/
{
cv::gpu::StereoBeliefPropagation d_bp(ndisp);
cv::gpu::GpuMat d_imgLeft(imgLeft);
cv::gpu::GpuMat d_imgRight(imgRight);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_imgLeft(imgLeft);
const cv::gpu::GpuMat d_imgRight(imgRight);
cv::gpu::GpuMat dst;
d_bp(d_imgLeft, d_imgRight, d_dst);
TEST_CYCLE() d_bp(d_imgLeft, d_imgRight, dst);
TEST_CYCLE()
{
d_bp(d_imgLeft, d_imgRight, d_dst);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL() << "No such CPU implementation analogy.";
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// StereoConstantSpaceBP
PERF_TEST_P(ImagePair, Calib3D_StereoConstantSpaceBP, Values(pair_string("gpu/stereobm/aloe-L.png", "gpu/stereobm/aloe-R.png")))
PERF_TEST_P(ImagePair, Calib3D_StereoConstantSpaceBP,
Values(pair_string("gpu/stereobm/aloe-L.png", "gpu/stereobm/aloe-R.png")))
{
declare.time(10.0);
@ -115,29 +101,25 @@ PERF_TEST_P(ImagePair, Calib3D_StereoConstantSpaceBP, Values(pair_string("gpu/st
{
cv::gpu::StereoConstantSpaceBP d_csbp(ndisp);
cv::gpu::GpuMat d_imgLeft(imgLeft);
cv::gpu::GpuMat d_imgRight(imgRight);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_imgLeft(imgLeft);
const cv::gpu::GpuMat d_imgRight(imgRight);
cv::gpu::GpuMat dst;
d_csbp(d_imgLeft, d_imgRight, d_dst);
TEST_CYCLE() d_csbp(d_imgLeft, d_imgRight, dst);
TEST_CYCLE()
{
d_csbp(d_imgLeft, d_imgRight, d_dst);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL() << "No such CPU implementation analogy.";
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// DisparityBilateralFilter
PERF_TEST_P(ImagePair, Calib3D_DisparityBilateralFilter, Values(pair_string("gpu/stereobm/aloe-L.png", "gpu/stereobm/aloe-disp.png")))
PERF_TEST_P(ImagePair, Calib3D_DisparityBilateralFilter,
Values(pair_string("gpu/stereobm/aloe-L.png", "gpu/stereobm/aloe-disp.png")))
{
const cv::Mat img = readImage(GET_PARAM(0), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
@ -151,22 +133,17 @@ PERF_TEST_P(ImagePair, Calib3D_DisparityBilateralFilter, Values(pair_string("gpu
{
cv::gpu::DisparityBilateralFilter d_filter(ndisp);
cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat d_disp(disp);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_img(img);
const cv::gpu::GpuMat d_disp(disp);
cv::gpu::GpuMat dst;
d_filter(d_disp, d_img, d_dst);
TEST_CYCLE() d_filter(d_disp, d_img, dst);
TEST_CYCLE()
{
d_filter(d_disp, d_img, d_dst);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL() << "No such CPU implementation analogy.";
FAIL_NO_CPU();
}
}
@ -175,45 +152,42 @@ PERF_TEST_P(ImagePair, Calib3D_DisparityBilateralFilter, Values(pair_string("gpu
DEF_PARAM_TEST_1(Count, int);
PERF_TEST_P(Count, Calib3D_TransformPoints, Values(5000, 10000, 20000))
PERF_TEST_P(Count, Calib3D_TransformPoints,
Values(5000, 10000, 20000))
{
const int count = GetParam();
cv::Mat src(1, count, CV_32FC3);
fillRandom(src, -100, 100);
declare.in(src, WARMUP_RNG);
const cv::Mat rvec = cv::Mat::ones(1, 3, CV_32FC1);
const cv::Mat tvec = cv::Mat::ones(1, 3, CV_32FC1);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::transformPoints(d_src, rvec, tvec, d_dst);
TEST_CYCLE() cv::gpu::transformPoints(d_src, rvec, tvec, dst);
TEST_CYCLE()
{
cv::gpu::transformPoints(d_src, rvec, tvec, d_dst);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL() << "No such CPU implementation analogy.";
FAIL_NO_CPU();
}
}
//////////////////////////////////////////////////////////////////////
// ProjectPoints
PERF_TEST_P(Count, Calib3D_ProjectPoints, Values(5000, 10000, 20000))
PERF_TEST_P(Count, Calib3D_ProjectPoints,
Values(5000, 10000, 20000))
{
const int count = GetParam();
cv::Mat src(1, count, CV_32FC3);
fillRandom(src, -100, 100);
declare.in(src, WARMUP_RNG);
const cv::Mat rvec = cv::Mat::ones(1, 3, CV_32FC1);
const cv::Mat tvec = cv::Mat::ones(1, 3, CV_32FC1);
@ -221,28 +195,18 @@ PERF_TEST_P(Count, Calib3D_ProjectPoints, Values(5000, 10000, 20000))
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::projectPoints(d_src, rvec, tvec, camera_mat, cv::Mat(), d_dst);
TEST_CYCLE() cv::gpu::projectPoints(d_src, rvec, tvec, camera_mat, cv::Mat(), dst);
TEST_CYCLE()
{
cv::gpu::projectPoints(d_src, rvec, tvec, camera_mat, cv::Mat(), d_dst);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::projectPoints(src, rvec, tvec, camera_mat, cv::noArray(), dst);
TEST_CYCLE()
{
cv::projectPoints(src, rvec, tvec, camera_mat, cv::noArray(), dst);
}
TEST_CYCLE() cv::projectPoints(src, rvec, tvec, camera_mat, cv::noArray(), dst);
CPU_SANITY_CHECK(dst);
}
@ -251,17 +215,18 @@ PERF_TEST_P(Count, Calib3D_ProjectPoints, Values(5000, 10000, 20000))
//////////////////////////////////////////////////////////////////////
// SolvePnPRansac
PERF_TEST_P(Count, Calib3D_SolvePnPRansac, Values(5000, 10000, 20000))
PERF_TEST_P(Count, Calib3D_SolvePnPRansac,
Values(5000, 10000, 20000))
{
declare.time(10.0);
const int count = GetParam();
cv::Mat object(1, count, CV_32FC3);
fillRandom(object, -100, 100);
declare.in(object, WARMUP_RNG);
cv::Mat camera_mat(3, 3, CV_32FC1);
fillRandom(camera_mat, 0.5, 1);
cv::randu(camera_mat, 0.5, 1);
camera_mat.at<float>(0, 1) = 0.f;
camera_mat.at<float>(1, 0) = 0.f;
camera_mat.at<float>(2, 0) = 0.f;
@ -269,79 +234,66 @@ PERF_TEST_P(Count, Calib3D_SolvePnPRansac, Values(5000, 10000, 20000))
const cv::Mat dist_coef(1, 8, CV_32F, cv::Scalar::all(0));
std::vector<cv::Point2f> image_vec;
cv::Mat rvec_gold(1, 3, CV_32FC1);
fillRandom(rvec_gold, 0, 1);
cv::randu(rvec_gold, 0, 1);
cv::Mat tvec_gold(1, 3, CV_32FC1);
fillRandom(tvec_gold, 0, 1);
cv::randu(tvec_gold, 0, 1);
std::vector<cv::Point2f> image_vec;
cv::projectPoints(object, rvec_gold, tvec_gold, camera_mat, dist_coef, image_vec);
cv::Mat image(1, count, CV_32FC2, &image_vec[0]);
const cv::Mat image(1, count, CV_32FC2, &image_vec[0]);
cv::Mat rvec;
cv::Mat tvec;
if (PERF_RUN_GPU())
{
cv::gpu::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
TEST_CYCLE() cv::gpu::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
TEST_CYCLE()
{
cv::gpu::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
}
GPU_SANITY_CHECK(rvec, 1e-3);
GPU_SANITY_CHECK(tvec, 1e-3);
}
else
{
cv::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
TEST_CYCLE() cv::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
TEST_CYCLE()
{
cv::solvePnPRansac(object, image, camera_mat, dist_coef, rvec, tvec);
}
CPU_SANITY_CHECK(rvec, 1e-6);
CPU_SANITY_CHECK(tvec, 1e-6);
}
CPU_SANITY_CHECK(rvec);
CPU_SANITY_CHECK(tvec);
}
//////////////////////////////////////////////////////////////////////
// ReprojectImageTo3D
PERF_TEST_P(Sz_Depth, Calib3D_ReprojectImageTo3D, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U, CV_16S)))
PERF_TEST_P(Sz_Depth, Calib3D_ReprojectImageTo3D,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16S)))
{
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
cv::Mat src(size, depth);
fillRandom(src, 5.0, 30.0);
declare.in(src, WARMUP_RNG);
cv::Mat Q(4, 4, CV_32FC1);
fillRandom(Q, 0.1, 1.0);
cv::randu(Q, 0.1, 1.0);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::reprojectImageTo3D(d_src, d_dst, Q);
TEST_CYCLE() cv::gpu::reprojectImageTo3D(d_src, dst, Q);
TEST_CYCLE()
{
cv::gpu::reprojectImageTo3D(d_src, d_dst, Q);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::reprojectImageTo3D(src, dst, Q);
TEST_CYCLE()
{
cv::reprojectImageTo3D(src, dst, Q);
}
TEST_CYCLE() cv::reprojectImageTo3D(src, dst, Q);
CPU_SANITY_CHECK(dst);
}
@ -350,32 +302,27 @@ PERF_TEST_P(Sz_Depth, Calib3D_ReprojectImageTo3D, Combine(GPU_TYPICAL_MAT_SIZES,
//////////////////////////////////////////////////////////////////////
// DrawColorDisp
PERF_TEST_P(Sz_Depth, Calib3D_DrawColorDisp, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U, CV_16S)))
PERF_TEST_P(Sz_Depth, Calib3D_DrawColorDisp,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16S)))
{
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
cv::Mat src(size, type);
fillRandom(src, 0, 255);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::drawColorDisp(d_src, d_dst, 255);
TEST_CYCLE() cv::gpu::drawColorDisp(d_src, dst, 255);
TEST_CYCLE()
{
cv::gpu::drawColorDisp(d_src, d_dst, 255);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL() << "No such CPU implementation analogy.";
FAIL_NO_CPU();
}
}
} // namespace

1330
modules/gpu/perf/perf_core.cpp
View File

File diff suppressed because it is too large Load Diff

172
modules/gpu/perf/perf_denoising.cpp
View File

@ -3,8 +3,7 @@
using namespace std;
using namespace testing;
#define GPU_DENOISING_IMAGE_SIZES testing::Values(perf::szVGA, perf::szXGA, perf::sz720p, perf::sz1080p)
#define GPU_DENOISING_IMAGE_SIZES testing::Values(perf::szVGA, perf::sz720p)
//////////////////////////////////////////////////////////////////////
// BilateralFilter
@ -12,96 +11,86 @@ using namespace testing;
DEF_PARAM_TEST(Sz_Depth_Cn_KernelSz, cv::Size, MatDepth, MatCn, int);
PERF_TEST_P(Sz_Depth_Cn_KernelSz, Denoising_BilateralFilter,
Combine(GPU_DENOISING_IMAGE_SIZES, Values(CV_8U, CV_32F), GPU_CHANNELS_1_3, Values(3, 5, 9)))
Combine(GPU_DENOISING_IMAGE_SIZES,
Values(CV_8U, CV_32F),
GPU_CHANNELS_1_3,
Values(3, 5, 9)))
{
declare.time(60.0);
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
int channels = GET_PARAM(2);
int kernel_size = GET_PARAM(3);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int kernel_size = GET_PARAM(3);
float sigma_color = 7;
float sigma_spatial = 5;
int borderMode = cv::BORDER_REFLECT101;
const float sigma_color = 7;
const float sigma_spatial = 5;
const int borderMode = cv::BORDER_REFLECT101;
int type = CV_MAKE_TYPE(depth, channels);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::bilateralFilter(d_src, d_dst, kernel_size, sigma_color, sigma_spatial, borderMode);
TEST_CYCLE() cv::gpu::bilateralFilter(d_src, dst, kernel_size, sigma_color, sigma_spatial, borderMode);
TEST_CYCLE()
{
cv::gpu::bilateralFilter(d_src, d_dst, kernel_size, sigma_color, sigma_spatial, borderMode);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::bilateralFilter(src, dst, kernel_size, sigma_color, sigma_spatial, borderMode);
TEST_CYCLE()
{
cv::bilateralFilter(src, dst, kernel_size, sigma_color, sigma_spatial, borderMode);
}
TEST_CYCLE() cv::bilateralFilter(src, dst, kernel_size, sigma_color, sigma_spatial, borderMode);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// nonLocalMeans
DEF_PARAM_TEST(Sz_Depth_Cn_WinSz_BlockSz, cv::Size, MatDepth, MatCn, int, int);
PERF_TEST_P(Sz_Depth_Cn_WinSz_BlockSz, Denoising_NonLocalMeans,
Combine(GPU_DENOISING_IMAGE_SIZES, Values<MatDepth>(CV_8U), GPU_CHANNELS_1_3, Values(21), Values(5, 7)))
Combine(GPU_DENOISING_IMAGE_SIZES,
Values<MatDepth>(CV_8U),
GPU_CHANNELS_1_3,
Values(21),
Values(5)))
{
declare.time(60.0);
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
int channels = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
const int search_widow_size = GET_PARAM(3);
const int block_size = GET_PARAM(4);
int search_widow_size = GET_PARAM(3);
int block_size = GET_PARAM(4);
const float h = 10;
const int borderMode = cv::BORDER_REFLECT101;
float h = 10;
int borderMode = cv::BORDER_REFLECT101;
int type = CV_MAKE_TYPE(depth, channels);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::nonLocalMeans(d_src, d_dst, h, search_widow_size, block_size, borderMode);
TEST_CYCLE() cv::gpu::nonLocalMeans(d_src, dst, h, search_widow_size, block_size, borderMode);
TEST_CYCLE()
{
cv::gpu::nonLocalMeans(d_src, d_dst, h, search_widow_size, block_size, borderMode);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
FAIL() << "No such CPU implementation analogy";
FAIL_NO_CPU();
}
}
@ -112,46 +101,41 @@ PERF_TEST_P(Sz_Depth_Cn_WinSz_BlockSz, Denoising_NonLocalMeans,
DEF_PARAM_TEST(Sz_Depth_Cn_WinSz_BlockSz, cv::Size, MatDepth, MatCn, int, int);
PERF_TEST_P(Sz_Depth_Cn_WinSz_BlockSz, Denoising_FastNonLocalMeans,
Combine(GPU_DENOISING_IMAGE_SIZES, Values<MatDepth>(CV_8U), GPU_CHANNELS_1_3, Values(21), Values(7)))
Combine(GPU_DENOISING_IMAGE_SIZES,
Values<MatDepth>(CV_8U),
GPU_CHANNELS_1_3,
Values(21),
Values(7)))
{
declare.time(150.0);
declare.time(60.0);
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int search_widow_size = GET_PARAM(2);
const int block_size = GET_PARAM(3);
int search_widow_size = GET_PARAM(2);
int block_size = GET_PARAM(3);
float h = 10;
int type = CV_MAKE_TYPE(depth, 1);
const float h = 10;
const int type = CV_MAKE_TYPE(depth, 1);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
cv::gpu::FastNonLocalMeansDenoising fnlmd;
fnlmd.simpleMethod(d_src, d_dst, h, search_widow_size, block_size);
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE()
{
fnlmd.simpleMethod(d_src, d_dst, h, search_widow_size, block_size);
}
TEST_CYCLE() fnlmd.simpleMethod(d_src, dst, h, search_widow_size, block_size);
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::fastNlMeansDenoising(src, dst, h, block_size, search_widow_size);
TEST_CYCLE()
{
cv::fastNlMeansDenoising(src, dst, h, block_size, search_widow_size);
}
TEST_CYCLE() cv::fastNlMeansDenoising(src, dst, h, block_size, search_widow_size);
CPU_SANITY_CHECK(dst);
}
@ -163,47 +147,41 @@ PERF_TEST_P(Sz_Depth_Cn_WinSz_BlockSz, Denoising_FastNonLocalMeans,
DEF_PARAM_TEST(Sz_Depth_WinSz_BlockSz, cv::Size, MatDepth, int, int);
PERF_TEST_P(Sz_Depth_WinSz_BlockSz, Denoising_FastNonLocalMeansColored,
Combine(GPU_DENOISING_IMAGE_SIZES, Values<MatDepth>(CV_8U), Values(21), Values(7)))
Combine(GPU_DENOISING_IMAGE_SIZES,
Values<MatDepth>(CV_8U),
Values(21),
Values(7)))
{
declare.time(350.0);
declare.time(60.0);
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int search_widow_size = GET_PARAM(2);
const int block_size = GET_PARAM(3);
int search_widow_size = GET_PARAM(2);
int block_size = GET_PARAM(3);
float h = 10;
int type = CV_MAKE_TYPE(depth, 3);
const float h = 10;
const int type = CV_MAKE_TYPE(depth, 3);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
cv::gpu::FastNonLocalMeansDenoising fnlmd;
fnlmd.labMethod(d_src, d_dst, h, h, search_widow_size, block_size);
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
TEST_CYCLE()
{
fnlmd.labMethod(d_src, d_dst, h, h, search_widow_size, block_size);
}
TEST_CYCLE() fnlmd.labMethod(d_src, dst, h, h, search_widow_size, block_size);
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::fastNlMeansDenoisingColored(src, dst, h, h, block_size, search_widow_size);
TEST_CYCLE()
{
cv::fastNlMeansDenoisingColored(src, dst, h, h, block_size, search_widow_size);
}
TEST_CYCLE() cv::fastNlMeansDenoisingColored(src, dst, h, h, block_size, search_widow_size);
CPU_SANITY_CHECK(dst);
}
}
}

328
modules/gpu/perf/perf_features2d.cpp
View File

@ -3,139 +3,194 @@
using namespace std;
using namespace testing;
namespace {
struct KeypointIdxCompare
{
std::vector<cv::KeyPoint>* keypoints;
explicit KeypointIdxCompare(std::vector<cv::KeyPoint>* _keypoints) : keypoints(_keypoints) {}
bool operator ()(size_t i1, size_t i2) const
{
cv::KeyPoint kp1 = (*keypoints)[i1];
cv::KeyPoint kp2 = (*keypoints)[i2];
if (kp1.pt.x != kp2.pt.x)
return kp1.pt.x < kp2.pt.x;
if (kp1.pt.y != kp2.pt.y)
return kp1.pt.y < kp2.pt.y;
if (kp1.response != kp2.response)
return kp1.response < kp2.response;
return kp1.octave < kp2.octave;
}
};
static void sortKeyPoints(std::vector<cv::KeyPoint>& keypoints, cv::InputOutputArray _descriptors = cv::noArray())
{
std::vector<size_t> indexies(keypoints.size());
for (size_t i = 0; i < indexies.size(); ++i)
indexies[i] = i;
std::sort(indexies.begin(), indexies.end(), KeypointIdxCompare(&keypoints));
std::vector<cv::KeyPoint> new_keypoints;
cv::Mat new_descriptors;
new_keypoints.resize(keypoints.size());
cv::Mat descriptors;
if (_descriptors.needed())
{
descriptors = _descriptors.getMat();
new_descriptors.create(descriptors.size(), descriptors.type());
}
for (size_t i = 0; i < indexies.size(); ++i)
{
size_t new_idx = indexies[i];
new_keypoints[i] = keypoints[new_idx];
if (!new_descriptors.empty())
descriptors.row((int) new_idx).copyTo(new_descriptors.row((int) i));
}
keypoints.swap(new_keypoints);
if (_descriptors.needed())
new_descriptors.copyTo(_descriptors);
}
//////////////////////////////////////////////////////////////////////
// SURF
DEF_PARAM_TEST_1(Image, string);
PERF_TEST_P(Image, Features2D_SURF, Values<string>("gpu/perf/aloe.png"))
PERF_TEST_P(Image, Features2D_SURF,
Values<string>("gpu/perf/aloe.png"))
{
declare.time(50.0);
cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
const cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
if (PERF_RUN_GPU())
{
cv::gpu::SURF_GPU d_surf;
cv::gpu::GpuMat d_img(img);
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat d_keypoints, d_descriptors;
d_surf(d_img, cv::gpu::GpuMat(), d_keypoints, d_descriptors);
TEST_CYCLE() d_surf(d_img, cv::gpu::GpuMat(), d_keypoints, d_descriptors);
TEST_CYCLE()
{
d_surf(d_img, cv::gpu::GpuMat(), d_keypoints, d_descriptors);
}
std::vector<cv::KeyPoint> gpu_keypoints;
d_surf.downloadKeypoints(d_keypoints, gpu_keypoints);
GPU_SANITY_CHECK(d_descriptors, 1e-4);
GPU_SANITY_CHECK_KEYPOINTS(SURF, d_keypoints);
cv::Mat gpu_descriptors(d_descriptors);
sortKeyPoints(gpu_keypoints, gpu_descriptors);
SANITY_CHECK_KEYPOINTS(gpu_keypoints);
SANITY_CHECK(gpu_descriptors);
}
else
{
cv::SURF surf;
std::vector<cv::KeyPoint> keypoints;
cv::Mat descriptors;
std::vector<cv::KeyPoint> cpu_keypoints;
cv::Mat cpu_descriptors;
surf(img, cv::noArray(), keypoints, descriptors);
TEST_CYCLE() surf(img, cv::noArray(), cpu_keypoints, cpu_descriptors);
TEST_CYCLE()
{
keypoints.clear();
surf(img, cv::noArray(), keypoints, descriptors);
}
SANITY_CHECK_KEYPOINTS(keypoints);
SANITY_CHECK(descriptors, 1e-4);
SANITY_CHECK_KEYPOINTS(cpu_keypoints);
SANITY_CHECK(cpu_descriptors);
}
}
//////////////////////////////////////////////////////////////////////
// FAST
PERF_TEST_P(Image, Features2D_FAST, Values<string>("gpu/perf/aloe.png"))
DEF_PARAM_TEST(Image_Threshold_NonMaxSupression, string, int, bool);
PERF_TEST_P(Image_Threshold_NonMaxSupression, Features2D_FAST,
Combine(Values<string>("gpu/perf/aloe.png"),
Values(20),
Bool()))
{
cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
const cv::Mat img = readImage(GET_PARAM(0), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
const int threshold = GET_PARAM(1);
const bool nonMaxSuppersion = GET_PARAM(2);
if (PERF_RUN_GPU())
{
cv::gpu::FAST_GPU d_fast(20);
cv::gpu::FAST_GPU d_fast(threshold, nonMaxSuppersion, 0.5);
cv::gpu::GpuMat d_img(img);
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat d_keypoints;
d_fast(d_img, cv::gpu::GpuMat(), d_keypoints);
TEST_CYCLE() d_fast(d_img, cv::gpu::GpuMat(), d_keypoints);
TEST_CYCLE()
{
d_fast(d_img, cv::gpu::GpuMat(), d_keypoints);
}
std::vector<cv::KeyPoint> gpu_keypoints;
d_fast.downloadKeypoints(d_keypoints, gpu_keypoints);
GPU_SANITY_CHECK_RESPONSE(FAST, d_keypoints);
sortKeyPoints(gpu_keypoints);
SANITY_CHECK_KEYPOINTS(gpu_keypoints);
}
else
{
std::vector<cv::KeyPoint> keypoints;
std::vector<cv::KeyPoint> cpu_keypoints;
cv::FAST(img, keypoints, 20);
TEST_CYCLE() cv::FAST(img, cpu_keypoints, threshold, nonMaxSuppersion);
TEST_CYCLE()
{
keypoints.clear();
cv::FAST(img, keypoints, 20);
}
SANITY_CHECK_KEYPOINTS(keypoints);
SANITY_CHECK_KEYPOINTS(cpu_keypoints);
}
}
//////////////////////////////////////////////////////////////////////
// ORB
PERF_TEST_P(Image, Features2D_ORB, Values<string>("gpu/perf/aloe.png"))
DEF_PARAM_TEST(Image_NFeatures, string, int);
PERF_TEST_P(Image_NFeatures, Features2D_ORB,
Combine(Values<string>("gpu/perf/aloe.png"),
Values(4000)))
{
cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
const cv::Mat img = readImage(GET_PARAM(0), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
const int nFeatures = GET_PARAM(1);
if (PERF_RUN_GPU())
{
cv::gpu::ORB_GPU d_orb(4000);
cv::gpu::ORB_GPU d_orb(nFeatures);
cv::gpu::GpuMat d_img(img);
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat d_keypoints, d_descriptors;
d_orb(d_img, cv::gpu::GpuMat(), d_keypoints, d_descriptors);
TEST_CYCLE() d_orb(d_img, cv::gpu::GpuMat(), d_keypoints, d_descriptors);
TEST_CYCLE()
{
d_orb(d_img, cv::gpu::GpuMat(), d_keypoints, d_descriptors);
}
std::vector<cv::KeyPoint> gpu_keypoints;
d_orb.downloadKeyPoints(d_keypoints, gpu_keypoints);
GPU_SANITY_CHECK_KEYPOINTS(ORB, d_keypoints);
GPU_SANITY_CHECK(d_descriptors);
cv::Mat gpu_descriptors(d_descriptors);
gpu_keypoints.resize(10);
gpu_descriptors = gpu_descriptors.rowRange(0, 10);
sortKeyPoints(gpu_keypoints, gpu_descriptors);
SANITY_CHECK_KEYPOINTS(gpu_keypoints);
SANITY_CHECK(gpu_descriptors);
}
else
{
cv::ORB orb(4000);
cv::ORB orb(nFeatures);
std::vector<cv::KeyPoint> keypoints;
cv::Mat descriptors;
std::vector<cv::KeyPoint> cpu_keypoints;
cv::Mat cpu_descriptors;
orb(img, cv::noArray(), keypoints, descriptors);
TEST_CYCLE() orb(img, cv::noArray(), cpu_keypoints, cpu_descriptors);
TEST_CYCLE()
{
keypoints.clear();
orb(img, cv::noArray(), keypoints, descriptors);
}
SANITY_CHECK_KEYPOINTS(keypoints);
SANITY_CHECK(descriptors);
SANITY_CHECK_KEYPOINTS(cpu_keypoints);
SANITY_CHECK(cpu_descriptors);
}
}
@ -144,166 +199,165 @@ PERF_TEST_P(Image, Features2D_ORB, Values<string>("gpu/perf/aloe.png"))
DEF_PARAM_TEST(DescSize_Norm, int, NormType);
PERF_TEST_P(DescSize_Norm, Features2D_BFMatch, Combine(Values(64, 128, 256), Values(NormType(cv::NORM_L1), NormType(cv::NORM_L2), NormType(cv::NORM_HAMMING))))
PERF_TEST_P(DescSize_Norm, Features2D_BFMatch,
Combine(Values(64, 128, 256),
Values(NormType(cv::NORM_L1), NormType(cv::NORM_L2), NormType(cv::NORM_HAMMING))))
{
declare.time(20.0);
int desc_size = GET_PARAM(0);
int normType = GET_PARAM(1);
const int desc_size = GET_PARAM(0);
const int normType = GET_PARAM(1);
int type = normType == cv::NORM_HAMMING ? CV_8U : CV_32F;
const int type = normType == cv::NORM_HAMMING ? CV_8U : CV_32F;
cv::Mat query(3000, desc_size, type);
fillRandom(query);
declare.in(query, WARMUP_RNG);
cv::Mat train(3000, desc_size, type);
fillRandom(train);
declare.in(train, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::BFMatcher_GPU d_matcher(normType);
cv::gpu::GpuMat d_query(query);
cv::gpu::GpuMat d_train(train);
const cv::gpu::GpuMat d_query(query);
const cv::gpu::GpuMat d_train(train);
cv::gpu::GpuMat d_trainIdx, d_distance;
d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
TEST_CYCLE() d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
TEST_CYCLE()
{
d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
}
std::vector<cv::DMatch> gpu_matches;
d_matcher.matchDownload(d_trainIdx, d_distance, gpu_matches);
GPU_SANITY_CHECK(d_trainIdx);
GPU_SANITY_CHECK(d_distance);
SANITY_CHECK_MATCHES(gpu_matches);
}
else
{
cv::BFMatcher matcher(normType);
std::vector<cv::DMatch> matches;
std::vector<cv::DMatch> cpu_matches;
matcher.match(query, train, matches);
TEST_CYCLE() matcher.match(query, train, cpu_matches);
TEST_CYCLE()
{
matcher.match(query, train, matches);
}
SANITY_CHECK(matches);
SANITY_CHECK_MATCHES(cpu_matches);
}
}
//////////////////////////////////////////////////////////////////////
// BFKnnMatch
static void toOneRowMatches(const std::vector< std::vector<cv::DMatch> >& src, std::vector<cv::DMatch>& dst)
{
dst.clear();
for (size_t i = 0; i < src.size(); ++i)
for (size_t j = 0; j < src[i].size(); ++j)
dst.push_back(src[i][j]);
}
DEF_PARAM_TEST(DescSize_K_Norm, int, int, NormType);
PERF_TEST_P(DescSize_K_Norm, Features2D_BFKnnMatch, Combine(
Values(64, 128, 256),
Values(2, 3),
Values(NormType(cv::NORM_L1), NormType(cv::NORM_L2), NormType(cv::NORM_HAMMING))))
PERF_TEST_P(DescSize_K_Norm, Features2D_BFKnnMatch,
Combine(Values(64, 128, 256),
Values(2, 3),
Values(NormType(cv::NORM_L1), NormType(cv::NORM_L2))))
{
declare.time(30.0);
int desc_size = GET_PARAM(0);
int k = GET_PARAM(1);
int normType = GET_PARAM(2);
const int desc_size = GET_PARAM(0);
const int k = GET_PARAM(1);
const int normType = GET_PARAM(2);
int type = normType == cv::NORM_HAMMING ? CV_8U : CV_32F;
const int type = normType == cv::NORM_HAMMING ? CV_8U : CV_32F;
cv::Mat query(3000, desc_size, type);
fillRandom(query);
declare.in(query, WARMUP_RNG);
cv::Mat train(3000, desc_size, type);
fillRandom(train);
declare.in(train, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::BFMatcher_GPU d_matcher(normType);
cv::gpu::GpuMat d_query(query);
cv::gpu::GpuMat d_train(train);
const cv::gpu::GpuMat d_query(query);
const cv::gpu::GpuMat d_train(train);
cv::gpu::GpuMat d_trainIdx, d_distance, d_allDist;
d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, k);
TEST_CYCLE() d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, k);
TEST_CYCLE()
{
d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, k);
}
std::vector< std::vector<cv::DMatch> > matchesTbl;
d_matcher.knnMatchDownload(d_trainIdx, d_distance, matchesTbl);
GPU_SANITY_CHECK(d_trainIdx);
GPU_SANITY_CHECK(d_distance);
std::vector<cv::DMatch> gpu_matches;
toOneRowMatches(matchesTbl, gpu_matches);
SANITY_CHECK_MATCHES(gpu_matches);
}
else
{
cv::BFMatcher matcher(normType);
std::vector< std::vector<cv::DMatch> > matches;
std::vector< std::vector<cv::DMatch> > matchesTbl;
matcher.knnMatch(query, train, matches, k);
TEST_CYCLE() matcher.knnMatch(query, train, matchesTbl, k);
TEST_CYCLE()
{
matcher.knnMatch(query, train, matches, k);
}
std::vector<cv::DMatch> cpu_matches;
toOneRowMatches(matchesTbl, cpu_matches);
SANITY_CHECK(matches);
SANITY_CHECK_MATCHES(cpu_matches);
}
}
//////////////////////////////////////////////////////////////////////
// BFRadiusMatch
PERF_TEST_P(DescSize_Norm, Features2D_BFRadiusMatch, Combine(Values(64, 128, 256), Values(NormType(cv::NORM_L1), NormType(cv::NORM_L2), NormType(cv::NORM_HAMMING))))
PERF_TEST_P(DescSize_Norm, Features2D_BFRadiusMatch,
Combine(Values(64, 128, 256),
Values(NormType(cv::NORM_L1), NormType(cv::NORM_L2))))
{
declare.time(30.0);
int desc_size = GET_PARAM(0);
int normType = GET_PARAM(1);
const int desc_size = GET_PARAM(0);
const int normType = GET_PARAM(1);
int type = normType == cv::NORM_HAMMING ? CV_8U : CV_32F;
const int type = normType == cv::NORM_HAMMING ? CV_8U : CV_32F;
const float maxDistance = 10000;
cv::Mat query(3000, desc_size, type);
fillRandom(query, 0.0, 1.0);
declare.in(query, WARMUP_RNG);
cv::Mat train(3000, desc_size, type);
fillRandom(train, 0.0, 1.0);
declare.in(train, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::BFMatcher_GPU d_matcher(normType);
cv::gpu::GpuMat d_query(query);
cv::gpu::GpuMat d_train(train);
const cv::gpu::GpuMat d_query(query);
const cv::gpu::GpuMat d_train(train);
cv::gpu::GpuMat d_trainIdx, d_nMatches, d_distance;
d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, 2.0);
TEST_CYCLE() d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, maxDistance);
TEST_CYCLE()
{
d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, 2.0);
}
std::vector< std::vector<cv::DMatch> > matchesTbl;
d_matcher.radiusMatchDownload(d_trainIdx, d_distance, d_nMatches, matchesTbl);
GPU_SANITY_CHECK(d_trainIdx);
GPU_SANITY_CHECK(d_distance);
std::vector<cv::DMatch> gpu_matches;
toOneRowMatches(matchesTbl, gpu_matches);
SANITY_CHECK_MATCHES(gpu_matches);
}
else
{
cv::BFMatcher matcher(normType);
std::vector< std::vector<cv::DMatch> > matches;
std::vector< std::vector<cv::DMatch> > matchesTbl;
matcher.radiusMatch(query, train, matches, 2.0);
TEST_CYCLE() matcher.radiusMatch(query, train, matchesTbl, maxDistance);
TEST_CYCLE()
{
matcher.radiusMatch(query, train, matches, 2.0);
}
std::vector<cv::DMatch> cpu_matches;
toOneRowMatches(matchesTbl, cpu_matches);
SANITY_CHECK(matches);
SANITY_CHECK_MATCHES(cpu_matches);
}
}
} // namespace

263
modules/gpu/perf/perf_filters.cpp
View File

@ -3,48 +3,39 @@
using namespace std;
using namespace testing;
namespace {
//////////////////////////////////////////////////////////////////////
// Blur
DEF_PARAM_TEST(Sz_Type_KernelSz, cv::Size, MatType, int);
PERF_TEST_P(Sz_Type_KernelSz, Filters_Blur, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8UC1, CV_8UC4), Values(3, 5, 7)))
PERF_TEST_P(Sz_Type_KernelSz, Filters_Blur,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8UC1, CV_8UC4),
Values(3, 5, 7)))
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
int ksize = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int ksize = GET_PARAM(2);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::blur(d_src, d_dst, cv::Size(ksize, ksize));
TEST_CYCLE() cv::gpu::blur(d_src, dst, cv::Size(ksize, ksize));
TEST_CYCLE()
{
cv::gpu::blur(d_src, d_dst, cv::Size(ksize, ksize));
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::blur(src, dst, cv::Size(ksize, ksize));
TEST_CYCLE()
{
cv::blur(src, dst, cv::Size(ksize, ksize));
}
TEST_CYCLE() cv::blur(src, dst, cv::Size(ksize, ksize));
CPU_SANITY_CHECK(dst);
}
@ -57,38 +48,28 @@ PERF_TEST_P(Sz_Type_KernelSz, Filters_Sobel, Combine(GPU_TYPICAL_MAT_SIZES, Valu
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
int ksize = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int ksize = GET_PARAM(2);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf;
cv::gpu::Sobel(d_src, d_dst, -1, 1, 1, d_buf, ksize);
TEST_CYCLE() cv::gpu::Sobel(d_src, dst, -1, 1, 1, d_buf, ksize);
TEST_CYCLE()
{
cv::gpu::Sobel(d_src, d_dst, -1, 1, 1, d_buf, ksize);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::Sobel(src, dst, -1, 1, 1, ksize);
TEST_CYCLE()
{
cv::Sobel(src, dst, -1, 1, 1, ksize);
}
TEST_CYCLE() cv::Sobel(src, dst, -1, 1, 1, ksize);
CPU_SANITY_CHECK(dst);
}
@ -101,37 +82,27 @@ PERF_TEST_P(Sz_Type, Filters_Scharr, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf;
cv::gpu::Scharr(d_src, d_dst, -1, 1, 0, d_buf);
TEST_CYCLE() cv::gpu::Scharr(d_src, dst, -1, 1, 0, d_buf);
TEST_CYCLE()
{
cv::gpu::Scharr(d_src, d_dst, -1, 1, 0, d_buf);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::Scharr(src, dst, -1, 1, 0);
TEST_CYCLE()
{
cv::Scharr(src, dst, -1, 1, 0);
}
TEST_CYCLE() cv::Scharr(src, dst, -1, 1, 0);
CPU_SANITY_CHECK(dst);
}
@ -144,38 +115,28 @@ PERF_TEST_P(Sz_Type_KernelSz, Filters_GaussianBlur, Combine(GPU_TYPICAL_MAT_SIZE
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
int ksize = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int ksize = GET_PARAM(2);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf;
cv::gpu::GaussianBlur(d_src, d_dst, cv::Size(ksize, ksize), d_buf, 0.5);
TEST_CYCLE() cv::gpu::GaussianBlur(d_src, dst, cv::Size(ksize, ksize), d_buf, 0.5);
TEST_CYCLE()
{
cv::gpu::GaussianBlur(d_src, d_dst, cv::Size(ksize, ksize), d_buf, 0.5);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::GaussianBlur(src, dst, cv::Size(ksize, ksize), 0.5);
TEST_CYCLE()
{
cv::GaussianBlur(src, dst, cv::Size(ksize, ksize), 0.5);
}
TEST_CYCLE() cv::GaussianBlur(src, dst, cv::Size(ksize, ksize), 0.5);
CPU_SANITY_CHECK(dst);
}
@ -188,37 +149,27 @@ PERF_TEST_P(Sz_Type_KernelSz, Filters_Laplacian, Combine(GPU_TYPICAL_MAT_SIZES,
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
int ksize = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int ksize = GET_PARAM(2);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::Laplacian(d_src, d_dst, -1, ksize);
TEST_CYCLE() cv::gpu::Laplacian(d_src, dst, -1, ksize);
TEST_CYCLE()
{
cv::gpu::Laplacian(d_src, d_dst, -1, ksize);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::Laplacian(src, dst, -1, ksize);
TEST_CYCLE()
{
cv::Laplacian(src, dst, -1, ksize);
}
TEST_CYCLE() cv::Laplacian(src, dst, -1, ksize);
CPU_SANITY_CHECK(dst);
}
@ -231,39 +182,29 @@ PERF_TEST_P(Sz_Type, Filters_Erode, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8UC
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
cv::Mat ker = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
const cv::Mat ker = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf;
cv::gpu::erode(d_src, d_dst, ker, d_buf);
TEST_CYCLE() cv::gpu::erode(d_src, dst, ker, d_buf);
TEST_CYCLE()
{
cv::gpu::erode(d_src, d_dst, ker, d_buf);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::erode(src, dst, ker);
TEST_CYCLE()
{
cv::erode(src, dst, ker);
}
TEST_CYCLE() cv::erode(src, dst, ker);
CPU_SANITY_CHECK(dst);
}
@ -276,39 +217,29 @@ PERF_TEST_P(Sz_Type, Filters_Dilate, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
cv::Mat ker = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
const cv::Mat ker = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf;
cv::gpu::dilate(d_src, d_dst, ker, d_buf);
TEST_CYCLE() cv::gpu::dilate(d_src, dst, ker, d_buf);
TEST_CYCLE()
{
cv::gpu::dilate(d_src, d_dst, ker, d_buf);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::dilate(src, dst, ker);
TEST_CYCLE()
{
cv::dilate(src, dst, ker);
}
TEST_CYCLE() cv::dilate(src, dst, ker);
CPU_SANITY_CHECK(dst);
}
@ -326,41 +257,31 @@ PERF_TEST_P(Sz_Type_Op, Filters_MorphologyEx, Combine(GPU_TYPICAL_MAT_SIZES, Val
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
int morphOp = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int morphOp = GET_PARAM(2);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
cv::Mat ker = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
const cv::Mat ker = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(3, 3));
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::GpuMat d_buf1;
cv::gpu::GpuMat d_buf2;
cv::gpu::morphologyEx(d_src, d_dst, morphOp, ker, d_buf1, d_buf2);
TEST_CYCLE() cv::gpu::morphologyEx(d_src, dst, morphOp, ker, d_buf1, d_buf2);
TEST_CYCLE()
{
cv::gpu::morphologyEx(d_src, d_dst, morphOp, ker, d_buf1, d_buf2);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::morphologyEx(src, dst, morphOp, ker);
TEST_CYCLE()
{
cv::morphologyEx(src, dst, morphOp, ker);
}
TEST_CYCLE() cv::morphologyEx(src, dst, morphOp, ker);
CPU_SANITY_CHECK(dst);
}
@ -373,43 +294,31 @@ PERF_TEST_P(Sz_Type_KernelSz, Filters_Filter2D, Combine(GPU_TYPICAL_MAT_SIZES, V
{
declare.time(20.0);
cv::Size size = GET_PARAM(0);
int type = GET_PARAM(1);
int ksize = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int type = GET_PARAM(1);
const int ksize = GET_PARAM(2);
cv::Mat src(size, type);
fillRandom(src);
declare.in(src, WARMUP_RNG);
cv::Mat kernel(ksize, ksize, CV_32FC1);
fillRandom(kernel, 0.0, 1.0);
declare.in(kernel, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
cv::gpu::filter2D(d_src, d_dst, -1, kernel);
TEST_CYCLE() cv::gpu::filter2D(d_src, dst, -1, kernel);
TEST_CYCLE()
{
cv::gpu::filter2D(d_src, d_dst, -1, kernel);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::filter2D(src, dst, -1, kernel);
TEST_CYCLE()
{
cv::filter2D(src, dst, -1, kernel);
}
TEST_CYCLE() cv::filter2D(src, dst, -1, kernel);
CPU_SANITY_CHECK(dst);
}
}
} // namespace

1591
modules/gpu/perf/perf_imgproc.cpp
View File

File diff suppressed because it is too large Load Diff

57
modules/gpu/perf/perf_labeling.cpp
View File

@ -3,8 +3,6 @@
using namespace std;
using namespace testing;
namespace {
DEF_PARAM_TEST_1(Image, string);
struct GreedyLabeling
@ -100,28 +98,45 @@ struct GreedyLabeling
dot* stack;
};
PERF_TEST_P(Image, Labeling_ConnectedComponents, Values<string>("gpu/labeling/aloe-disp.png"))
PERF_TEST_P(Image, Labeling_ConnectivityMask,
Values<string>("gpu/labeling/aloe-disp.png"))
{
declare.time(1.0);
cv::Mat image = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
const cv::Mat image = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_image(image);
cv::gpu::GpuMat mask;
mask.create(image.rows, image.cols, CV_8UC1);
TEST_CYCLE() cv::gpu::connectivityMask(d_image, mask, cv::Scalar::all(0), cv::Scalar::all(2));
GPU_SANITY_CHECK(mask);
}
else
{
FAIL_NO_CPU();
}
}
PERF_TEST_P(Image, Labeling_ConnectedComponents,
Values<string>("gpu/labeling/aloe-disp.png"))
{
declare.time(1.0);
const cv::Mat image = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(image.empty());
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_mask;
cv::gpu::connectivityMask(cv::gpu::GpuMat(image), d_mask, cv::Scalar::all(0), cv::Scalar::all(2));
cv::gpu::GpuMat components;
components.create(image.rows, image.cols, CV_32SC1);
cv::gpu::connectivityMask(cv::gpu::GpuMat(image), mask, cv::Scalar::all(0), cv::Scalar::all(2));
ASSERT_NO_THROW(cv::gpu::labelComponents(mask, components));
TEST_CYCLE()
{
cv::gpu::labelComponents(mask, components);
}
TEST_CYCLE() cv::gpu::labelComponents(d_mask, components);
GPU_SANITY_CHECK(components);
}
@ -129,17 +144,9 @@ PERF_TEST_P(Image, Labeling_ConnectedComponents, Values<string>("gpu/labeling/al
{
GreedyLabeling host(image);
host(host._labels);
TEST_CYCLE() host(host._labels);
declare.time(1.0);
TEST_CYCLE()
{
host(host._labels);
}
CPU_SANITY_CHECK(host._labels);
cv::Mat components = host._labels;
CPU_SANITY_CHECK(components);
}
}
} // namespace

6
modules/gpu/perf/perf_main.cpp
View File

@ -1,7 +1,5 @@
#include "perf_precomp.hpp"
namespace{
static void printOsInfo()
{
#if defined _WIN32
@ -69,6 +67,4 @@ static void printCudaInfo()
#endif
}
}
CV_PERF_TEST_MAIN(gpu, printCudaInfo())
CV_PERF_TEST_MAIN(gpu, printCudaInfo())

159
modules/gpu/perf/perf_matop.cpp
View File

@ -3,137 +3,112 @@
using namespace std;
using namespace testing;
namespace {
//////////////////////////////////////////////////////////////////////
// SetTo
PERF_TEST_P(Sz_Depth_Cn, MatOp_SetTo, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U, CV_16U, CV_32F, CV_64F), GPU_CHANNELS_1_3_4))
PERF_TEST_P(Sz_Depth_Cn, MatOp_SetTo,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
GPU_CHANNELS_1_3_4))
{
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
int channels = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
int type = CV_MAKE_TYPE(depth, channels);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Scalar val(1, 2, 3, 4);
const cv::Scalar val(1, 2, 3, 4);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(size, type);
cv::gpu::GpuMat dst(size, type);
d_src.setTo(val);
TEST_CYCLE() dst.setTo(val);
TEST_CYCLE()
{
d_src.setTo(val);
}
GPU_SANITY_CHECK(d_src);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat src(size, type);
cv::Mat dst(size, type);
src.setTo(val);
TEST_CYCLE() dst.setTo(val);
TEST_CYCLE()
{
src.setTo(val);
}
CPU_SANITY_CHECK(src);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// SetToMasked
PERF_TEST_P(Sz_Depth_Cn, MatOp_SetToMasked, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U, CV_16U, CV_32F, CV_64F), GPU_CHANNELS_1_3_4))
PERF_TEST_P(Sz_Depth_Cn, MatOp_SetToMasked,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
GPU_CHANNELS_1_3_4))
{
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
int channels = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
int type = CV_MAKE_TYPE(depth, channels);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
fillRandom(src);
cv::Mat mask(size, CV_8UC1);
fillRandom(mask, 0, 2);
declare.in(src, mask, WARMUP_RNG);
cv::Scalar val(1, 2, 3, 4);
const cv::Scalar val(1, 2, 3, 4);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_mask(mask);
cv::gpu::GpuMat dst(src);
const cv::gpu::GpuMat d_mask(mask);
d_src.setTo(val, d_mask);
TEST_CYCLE() dst.setTo(val, d_mask);
TEST_CYCLE()
{
d_src.setTo(val, d_mask);
}
GPU_SANITY_CHECK(d_src);
GPU_SANITY_CHECK(dst);
}
else
{
src.setTo(val, mask);
cv::Mat dst = src;
TEST_CYCLE()
{
src.setTo(val, mask);
}
TEST_CYCLE() dst.setTo(val, mask);
CPU_SANITY_CHECK(src);
CPU_SANITY_CHECK(dst);
}
}
//////////////////////////////////////////////////////////////////////
// CopyToMasked
PERF_TEST_P(Sz_Depth_Cn, MatOp_CopyToMasked, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U, CV_16U, CV_32F, CV_64F), GPU_CHANNELS_1_3_4))
PERF_TEST_P(Sz_Depth_Cn, MatOp_CopyToMasked,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
GPU_CHANNELS_1_3_4))
{
cv::Size size = GET_PARAM(0);
int depth = GET_PARAM(1);
int channels = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int depth = GET_PARAM(1);
const int channels = GET_PARAM(2);
int type = CV_MAKE_TYPE(depth, channels);
const int type = CV_MAKE_TYPE(depth, channels);
cv::Mat src(size, type);
fillRandom(src);
cv::Mat mask(size, CV_8UC1);
fillRandom(mask, 0, 2);
declare.in(src, mask, WARMUP_RNG);
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_mask(mask);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
const cv::gpu::GpuMat d_mask(mask);
cv::gpu::GpuMat dst(d_src.size(), d_src.type(), cv::Scalar::all(0));
d_src.copyTo(d_dst, d_mask);
TEST_CYCLE() d_src.copyTo(dst, d_mask);
TEST_CYCLE()
{
d_src.copyTo(d_dst, d_mask);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
cv::Mat dst(src.size(), src.type(), cv::Scalar::all(0));
src.copyTo(dst, mask);
TEST_CYCLE()
{
src.copyTo(dst, mask);
}
TEST_CYCLE() src.copyTo(dst, mask);
CPU_SANITY_CHECK(dst);
}
@ -144,42 +119,36 @@ PERF_TEST_P(Sz_Depth_Cn, MatOp_CopyToMasked, Combine(GPU_TYPICAL_MAT_SIZES, Valu
DEF_PARAM_TEST(Sz_2Depth, cv::Size, MatDepth, MatDepth);
PERF_TEST_P(Sz_2Depth, MatOp_ConvertTo, Combine(GPU_TYPICAL_MAT_SIZES, Values(CV_8U, CV_16U, CV_32F, CV_64F), Values(CV_8U, CV_16U, CV_32F, CV_64F)))
PERF_TEST_P(Sz_2Depth, MatOp_ConvertTo,
Combine(GPU_TYPICAL_MAT_SIZES,
Values(CV_8U, CV_16U, CV_32F, CV_64F),
Values(CV_8U, CV_16U, CV_32F, CV_64F)))
{
cv::Size size = GET_PARAM(0);
int depth1 = GET_PARAM(1);
int depth2 = GET_PARAM(2);
const cv::Size size = GET_PARAM(0);
const int depth1 = GET_PARAM(1);
const int depth2 = GET_PARAM(2);
cv::Mat src(size, depth1);
fillRandom(src);
declare.in(src, WARMUP_RNG);
const double a = 0.5;
const double b = 1.0;
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat d_dst;
const cv::gpu::GpuMat d_src(src);
cv::gpu::GpuMat dst;
d_src.convertTo(d_dst, depth2, 0.5, 1.0);
TEST_CYCLE() d_src.convertTo(dst, depth2, a, b);
TEST_CYCLE()
{
d_src.convertTo(d_dst, depth2, 0.5, 1.0);
}
GPU_SANITY_CHECK(d_dst);
GPU_SANITY_CHECK(dst);
}
else
{
cv::Mat dst;
src.convertTo(dst, depth2, 0.5, 1.0);
TEST_CYCLE()
{
src.convertTo(dst, depth2, 0.5, 1.0);
}
TEST_CYCLE() src.convertTo(dst, depth2, a, b);
CPU_SANITY_CHECK(dst);
}
}
} // namespace

147
modules/gpu/perf/perf_objdetect.cpp
View File

@ -3,90 +3,47 @@
using namespace std;
using namespace testing;
namespace {
///////////////////////////////////////////////////////////////
// HOG
DEF_PARAM_TEST_1(Image, string);
PERF_TEST_P(Image, ObjDetect_HOG, Values<string>("gpu/hog/road.png"))
PERF_TEST_P(Image, ObjDetect_HOG,
Values<string>("gpu/hog/road.png",
"gpu/caltech/image_00000009_0.png",
"gpu/caltech/image_00000032_0.png",
"gpu/caltech/image_00000165_0.png",
"gpu/caltech/image_00000261_0.png",
"gpu/caltech/image_00000469_0.png",
"gpu/caltech/image_00000527_0.png",
"gpu/caltech/image_00000574_0.png"))
{
cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
const cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
std::vector<cv::Rect> found_locations;
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_img(img);
const cv::gpu::GpuMat d_img(img);
std::vector<cv::Rect> gpu_found_locations;
cv::gpu::HOGDescriptor d_hog;
d_hog.setSVMDetector(cv::gpu::HOGDescriptor::getDefaultPeopleDetector());
d_hog.detectMultiScale(d_img, found_locations);
TEST_CYCLE() d_hog.detectMultiScale(d_img, gpu_found_locations);
TEST_CYCLE()
{
d_hog.detectMultiScale(d_img, found_locations);
}
SANITY_CHECK(gpu_found_locations);
}
else
{
std::vector<cv::Rect> cpu_found_locations;
cv::HOGDescriptor hog;
hog.setSVMDetector(cv::gpu::HOGDescriptor::getDefaultPeopleDetector());
hog.detectMultiScale(img, found_locations);
TEST_CYCLE() hog.detectMultiScale(img, cpu_found_locations);
TEST_CYCLE()
{
hog.detectMultiScale(img, found_locations);
}
SANITY_CHECK(cpu_found_locations);
}
SANITY_CHECK(found_locations);
}
//===========test for CalTech data =============//
DEF_PARAM_TEST_1(HOG, string);
PERF_TEST_P(HOG, CalTech, Values<string>("gpu/caltech/image_00000009_0.png", "gpu/caltech/image_00000032_0.png",
"gpu/caltech/image_00000165_0.png", "gpu/caltech/image_00000261_0.png", "gpu/caltech/image_00000469_0.png",
"gpu/caltech/image_00000527_0.png", "gpu/caltech/image_00000574_0.png"))
{
cv::Mat img = readImage(GetParam(), cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
std::vector<cv::Rect> found_locations;
if (PERF_RUN_GPU())
{
cv::gpu::GpuMat d_img(img);
cv::gpu::HOGDescriptor d_hog;
d_hog.setSVMDetector(cv::gpu::HOGDescriptor::getDefaultPeopleDetector());
d_hog.detectMultiScale(d_img, found_locations);
TEST_CYCLE()
{
d_hog.detectMultiScale(d_img, found_locations);
}
}
else
{
cv::HOGDescriptor hog;
hog.setSVMDetector(cv::gpu::HOGDescriptor::getDefaultPeopleDetector());
hog.detectMultiScale(img, found_locations);
TEST_CYCLE()
{
hog.detectMultiScale(img, found_locations);
}
}
SANITY_CHECK(found_locations);
}
///////////////////////////////////////////////////////////////
@ -96,9 +53,9 @@ typedef pair<string, string> pair_string;
DEF_PARAM_TEST_1(ImageAndCascade, pair_string);
PERF_TEST_P(ImageAndCascade, ObjDetect_HaarClassifier,
Values<pair_string>(make_pair("gpu/haarcascade/group_1_640x480_VGA.pgm", "gpu/perf/haarcascade_frontalface_alt.xml")))
Values<pair_string>(make_pair("gpu/haarcascade/group_1_640x480_VGA.pgm", "gpu/perf/haarcascade_frontalface_alt.xml")))
{
cv::Mat img = readImage(GetParam().first, cv::IMREAD_GRAYSCALE);
const cv::Mat img = readImage(GetParam().first, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
if (PERF_RUN_GPU())
@ -106,33 +63,28 @@ PERF_TEST_P(ImageAndCascade, ObjDetect_HaarClassifier,
cv::gpu::CascadeClassifier_GPU d_cascade;
ASSERT_TRUE(d_cascade.load(perf::TestBase::getDataPath(GetParam().second)));
cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat d_objects_buffer;
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat objects_buffer;
int detections_num = 0;
d_cascade.detectMultiScale(d_img, d_objects_buffer);
TEST_CYCLE() detections_num = d_cascade.detectMultiScale(d_img, objects_buffer);
TEST_CYCLE()
{
d_cascade.detectMultiScale(d_img, d_objects_buffer);
}
GPU_SANITY_CHECK(d_objects_buffer);
std::vector<cv::Rect> gpu_rects(detections_num);
cv::Mat gpu_rects_mat(1, detections_num, cv::DataType<cv::Rect>::type, &gpu_rects[0]);
objects_buffer.colRange(0, detections_num).download(gpu_rects_mat);
cv::groupRectangles(gpu_rects, 3, 0.2);
SANITY_CHECK(gpu_rects);
}
else
{
cv::CascadeClassifier cascade;
ASSERT_TRUE(cascade.load(perf::TestBase::getDataPath("gpu/perf/haarcascade_frontalface_alt.xml")));
std::vector<cv::Rect> rects;
std::vector<cv::Rect> cpu_rects;
cascade.detectMultiScale(img, rects);
TEST_CYCLE() cascade.detectMultiScale(img, cpu_rects);
TEST_CYCLE()
{
cascade.detectMultiScale(img, rects);
}
CPU_SANITY_CHECK(rects);
SANITY_CHECK(cpu_rects);
}
}
@ -140,9 +92,9 @@ PERF_TEST_P(ImageAndCascade, ObjDetect_HaarClassifier,
// LBP cascade
PERF_TEST_P(ImageAndCascade, ObjDetect_LBPClassifier,
Values<pair_string>(make_pair("gpu/haarcascade/group_1_640x480_VGA.pgm", "gpu/lbpcascade/lbpcascade_frontalface.xml")))
Values<pair_string>(make_pair("gpu/haarcascade/group_1_640x480_VGA.pgm", "gpu/lbpcascade/lbpcascade_frontalface.xml")))
{
cv::Mat img = readImage(GetParam().first, cv::IMREAD_GRAYSCALE);
const cv::Mat img = readImage(GetParam().first, cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
if (PERF_RUN_GPU())
@ -150,34 +102,27 @@ PERF_TEST_P(ImageAndCascade, ObjDetect_LBPClassifier,
cv::gpu::CascadeClassifier_GPU d_cascade;
ASSERT_TRUE(d_cascade.load(perf::TestBase::getDataPath(GetParam().second)));
cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat d_gpu_rects;
const cv::gpu::GpuMat d_img(img);
cv::gpu::GpuMat objects_buffer;
int detections_num = 0;
d_cascade.detectMultiScale(d_img, d_gpu_rects);
TEST_CYCLE() detections_num = d_cascade.detectMultiScale(d_img, objects_buffer);
TEST_CYCLE()
{
d_cascade.detectMultiScale(d_img, d_gpu_rects);
}
GPU_SANITY_CHECK(d_gpu_rects);
std::vector<cv::Rect> gpu_rects(detections_num);
cv::Mat gpu_rects_mat(1, detections_num, cv::DataType<cv::Rect>::type, &gpu_rects[0]);
objects_buffer.colRange(0, detections_num).download(gpu_rects_mat);
cv::groupRectangles(gpu_rects, 3, 0.2);
SANITY_CHECK(gpu_rects);
}
else
{
cv::CascadeClassifier cascade;
ASSERT_TRUE(cascade.load(perf::TestBase::getDataPath("gpu/lbpcascade/lbpcascade_frontalface.xml")));
std::vector<cv::Rect> rects;
std::vector<cv::Rect> cpu_rects;
cascade.detectMultiScale(img, rects);
TEST_CYCLE() cascade.detectMultiScale(img, cpu_rects);
TEST_CYCLE()
{
cascade.detectMultiScale(img, rects);
}
CPU_SANITY_CHECK(rects);
SANITY_CHECK(cpu_rects);
}
}
} // namespace

766
modules/gpu/perf/perf_video.cpp
View File

File diff suppressed because it is too large Load Diff

9
modules/gpu/perf/utility.cpp
View File

@ -2,13 +2,6 @@
using namespace std;
using namespace cv;
using namespace cv::gpu;
void fillRandom(Mat& m, double a, double b)
{
RNG rng(123456789);
rng.fill(m, RNG::UNIFORM, Scalar::all(a), Scalar::all(b));
}
Mat readImage(const string& fileName, int flags)
{
@ -188,4 +181,4 @@ void PrintTo(const CvtColorInfo& info, ostream* os)
};
*os << str[info.code];
}
}

49
modules/gpu/perf/utility.hpp
View File

@ -2,11 +2,9 @@
#define __OPENCV_PERF_GPU_UTILITY_HPP__
#include "opencv2/core/core.hpp"
#include "opencv2/core/gpumat.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/ts/ts_perf.hpp"
void fillRandom(cv::Mat& m, double a = 0.0, double b = 255.0);
cv::Mat readImage(const std::string& fileName, int flags = cv::IMREAD_COLOR);
using perf::MatType;
@ -17,12 +15,13 @@ CV_ENUM(BorderMode, cv::BORDER_REFLECT101, cv::BORDER_REPLICATE, cv::BORDER_CONS
CV_ENUM(Interpolation, cv::INTER_NEAREST, cv::INTER_LINEAR, cv::INTER_CUBIC, cv::INTER_AREA)
#define ALL_INTERPOLATIONS testing::ValuesIn(Interpolation::all())
CV_ENUM(NormType, cv::NORM_INF, cv::NORM_L1, cv::NORM_L2, cv::NORM_HAMMING, cv::NORM_MINMAX)
const int Gray = 1, TwoChannel = 2, BGR = 3, BGRA = 4;
enum { Gray = 1, TwoChannel = 2, BGR = 3, BGRA = 4 };
CV_ENUM(MatCn, Gray, TwoChannel, BGR, BGRA)
#define GPU_CHANNELS_1_3_4 testing::Values(Gray, BGR, BGRA)
#define GPU_CHANNELS_1_3 testing::Values(Gray, BGR)
#define GPU_CHANNELS_1_3_4 testing::Values(MatCn(Gray), MatCn(BGR), MatCn(BGRA))
#define GPU_CHANNELS_1_3 testing::Values(MatCn(Gray), MatCn(BGR))
struct CvtColorInfo
{
@ -30,7 +29,8 @@ struct CvtColorInfo
int dcn;
int code;
explicit CvtColorInfo(int scn_=0, int dcn_=0, int code_=0) : scn(scn_), dcn(dcn_), code(code_) {}
CvtColorInfo() {}
explicit CvtColorInfo(int scn_, int dcn_, int code_) : scn(scn_), dcn(dcn_), code(code_) {}
};
void PrintTo(const CvtColorInfo& info, std::ostream* os);
@ -46,39 +46,18 @@ DEF_PARAM_TEST(Sz_Depth_Cn, cv::Size, MatDepth, MatCn);
#define GPU_TYPICAL_MAT_SIZES testing::Values(perf::sz720p, perf::szSXGA, perf::sz1080p)
#define GPU_SANITY_CHECK(dmat, ...) \
#define FAIL_NO_CPU() FAIL() << "No such CPU implementation analogy"
#define GPU_SANITY_CHECK(mat, ...) \
do{ \
cv::Mat d##dmat(dmat); \
SANITY_CHECK(d##dmat, ## __VA_ARGS__); \
cv::Mat gpu_##mat(mat); \
SANITY_CHECK(gpu_##mat, ## __VA_ARGS__); \
} while(0)
#define CPU_SANITY_CHECK(cmat, ...) \
#define CPU_SANITY_CHECK(mat, ...) \
do{ \
SANITY_CHECK(cmat, ## __VA_ARGS__); \
cv::Mat cpu_##mat(mat); \
SANITY_CHECK(cpu_##mat, ## __VA_ARGS__); \
} while(0)
#define GPU_SANITY_CHECK_KEYPOINTS(alg, dmat, ...) \
do{ \
cv::Mat d##dmat(dmat); \
cv::Mat __pt_x = d##dmat.row(cv::gpu::alg##_GPU::X_ROW); \
cv::Mat __pt_y = d##dmat.row(cv::gpu::alg##_GPU::Y_ROW); \
cv::Mat __angle = d##dmat.row(cv::gpu::alg##_GPU::ANGLE_ROW); \
cv::Mat __octave = d##dmat.row(cv::gpu::alg##_GPU::OCTAVE_ROW); \
cv::Mat __size = d##dmat.row(cv::gpu::alg##_GPU::SIZE_ROW); \
::perf::Regression::add(this, std::string(#dmat) + "-pt-x-row", __pt_x, ## __VA_ARGS__); \
::perf::Regression::add(this, std::string(#dmat) + "-pt-y-row", __pt_y, ## __VA_ARGS__); \
::perf::Regression::add(this, std::string(#dmat) + "-angle-row", __angle, ## __VA_ARGS__); \
::perf::Regression::add(this, std::string(#dmat) + "octave-row", __octave, ## __VA_ARGS__); \
::perf::Regression::add(this, std::string(#dmat) + "-pt-size-row", __size, ## __VA_ARGS__); \
} while(0)
#define GPU_SANITY_CHECK_RESPONSE(alg, dmat, ...) \
do{ \
cv::Mat d##dmat(dmat); \
cv::Mat __response = d##dmat.row(cv::gpu::alg##_GPU::RESPONSE_ROW); \
::perf::Regression::add(this, std::string(#dmat) + "-response-row", __response, ## __VA_ARGS__); \
} while(0)
#define FAIL_NO_CPU() FAIL() << "No such CPU implementation analogy"
#endif // __OPENCV_PERF_GPU_UTILITY_HPP__

8
modules/gpu/src/graphcuts.cpp
View File

@ -104,12 +104,12 @@ void cv::gpu::connectivityMask(const GpuMat& image, GpuMat& mask, const cv::Scal
void cv::gpu::labelComponents(const GpuMat& mask, GpuMat& components, int flags, Stream& s)
{
if (!TargetArchs::builtWith(SHARED_ATOMICS) || !DeviceInfo().supports(SHARED_ATOMICS))
CV_Error(CV_StsNotImplemented, "The device doesn't support shared atomics and communicative synchronization!");
CV_Assert(!mask.empty() && mask.type() == CV_8U);
if (mask.size() != components.size() || components.type() != CV_32SC1)
components.create(mask.size(), CV_32SC1);
if (!deviceSupports(SHARED_ATOMICS))
CV_Error(CV_StsNotImplemented, "The device doesn't support shared atomics and communicative synchronization!");
components.create(mask.size(), CV_32SC1);
cudaStream_t stream = StreamAccessor::getStream(s);
device::ccl::labelComponents(mask, components, flags, stream);

1
modules/gpu/src/imgproc.cpp
View File

@ -522,6 +522,7 @@ void cv::gpu::rotate(const GpuMat& src, GpuMat& dst, Size dsize, double angle, d
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
dst.create(dsize, src.type());
dst.setTo(Scalar::all(0));
funcs[src.depth()][src.channels() - 1](src, dst, dsize, angle, xShift, yShift, interpolation, StreamAccessor::getStream(stream));
}

1
modules/gpu/src/mssegmentation.cpp
View File

@ -382,6 +382,7 @@ void cv::gpu::meanShiftSegmentation(const GpuMat& src, Mat& dst, int sp, int sr,
dstcol[0] = static_cast<uchar>(sumcol[0] / comps.size[parent]);
dstcol[1] = static_cast<uchar>(sumcol[1] / comps.size[parent]);
dstcol[2] = static_cast<uchar>(sumcol[2] / comps.size[parent]);
dstcol[3] = 255;
}
}
}

2
modules/gpu/src/pyrlk.cpp
View File

@ -209,6 +209,8 @@ void cv::gpu::PyrLKOpticalFlow::dense(const GpuMat& prevImg, const GpuMat& nextI
ensureSizeIsEnough(prevImg.size(), CV_32FC1, vPyr_[0]);
ensureSizeIsEnough(prevImg.size(), CV_32FC1, uPyr_[1]);
ensureSizeIsEnough(prevImg.size(), CV_32FC1, vPyr_[1]);
uPyr_[0].setTo(Scalar::all(0));
vPyr_[0].setTo(Scalar::all(0));
uPyr_[1].setTo(Scalar::all(0));
vPyr_[1].setTo(Scalar::all(0));

12
modules/gpu/src/warp.cpp
View File

@ -232,10 +232,8 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
};
bool useNpp = borderMode == BORDER_CONSTANT && ofs.x == 0 && ofs.y == 0 && useNppTab[src.depth()][src.channels() - 1][interpolation];
#ifdef linux
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
#endif
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
if (useNpp)
{
@ -372,10 +370,8 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
};
bool useNpp = borderMode == BORDER_CONSTANT && ofs.x == 0 && ofs.y == 0 && useNppTab[src.depth()][src.channels() - 1][interpolation];
#ifdef linux
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
#endif
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
if (useNpp)
{