Merge pull request #13221 from elatkin:el/gapi_perf_sepfilter

GAPI (fluid): optimization of Separable filter (#13221)

* GAPI (fluid): Separable filter: performance test

* GAPI (fluid): enable all performance tests

* GAPI: separable filters: alternative code for Sobel

* GAPI (fluid): hide unused old code for Sobel filter

* GAPI (fluid): especial code for Sobel if U8 into S16

* GAPI (fluid): back to old code for Sobel

* GAPI (fluid): run_sepfilter3x3_impl() with CPU dispatcher

* GAPI (fluid): run_sepfilter3x3_impl(): fix compiler warnings

* GAPI (fluid): new engine for separable filters (but Sobel)

* GAPI (fluid): new performance engine for Sobel

* GAPI (fluid): Sepfilters performance: fixed compilation error
This commit is contained in:
Evgeny Latkin
2018-11-26 15:05:35 +03:00
committed by Alexander Alekhin
parent dd952f6d68
commit f07856eab9
6 changed files with 902 additions and 362 deletions
@@ -52,7 +52,7 @@ PERF_TEST_P_(SepFilterPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -100,7 +100,7 @@ PERF_TEST_P_(Filter2DPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -145,7 +145,7 @@ PERF_TEST_P_(BoxFilterPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -188,7 +188,7 @@ PERF_TEST_P_(BlurPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -230,7 +230,7 @@ PERF_TEST_P_(GaussianBlurPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -271,7 +271,7 @@ PERF_TEST_P_(MedianBlurPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -314,7 +314,7 @@ PERF_TEST_P_(ErodePerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -357,7 +357,7 @@ PERF_TEST_P_(Erode3x3PerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -400,7 +400,7 @@ PERF_TEST_P_(DilatePerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -443,7 +443,7 @@ PERF_TEST_P_(Dilate3x3PerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -526,7 +526,7 @@ PERF_TEST_P_(CannyPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -564,7 +564,7 @@ PERF_TEST_P_(EqHistPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -830,7 +830,7 @@ PERF_TEST_P_(LUV2BGRPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -13,9 +13,101 @@
namespace opencv_test
{
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid, SobelPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1), // add CV_32FC1 when ready
INSTANTIATE_TEST_CASE_P(SepFilterPerfTestFluid_8U, SepFilterPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3),
Values(3),
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_16S, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SepFilterPerfTestFluid_other, SepFilterPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_16UC1, CV_16SC1, CV_32FC1),
Values(3),
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(Filter2DPerfTestFluid, Filter2DPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add 4, 5, 7 when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::BORDER_DEFAULT),
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BoxFilterPerfTestFluid, BoxFilterPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::BORDER_DEFAULT),
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BlurPerfTestFluid, BlurPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::BORDER_DEFAULT),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(GaussianBlurPerfTestFluid, GaussianBlurPerfTest,
Combine(Values(ToleranceFilter(1e-3f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(MedianBlurPerfTestFluid, MedianBlurPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(ErodePerfTestFluid, ErodePerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::MorphShapes::MORPH_RECT,
cv::MorphShapes::MORPH_CROSS,
cv::MorphShapes::MORPH_ELLIPSE),
Values(cv::compile_args(IMGPROC_FLUID))));
// GAPI/fluid does not support iterations parameter for the Erode kernel
INSTANTIATE_TEST_CASE_P(DISABLED_Erode3x3PerfTestFluid, Erode3x3PerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(szVGA, sz720p, sz1080p),
Values(1, 2, 4),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(DilatePerfTestFluid, DilatePerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::MorphShapes::MORPH_RECT,
cv::MorphShapes::MORPH_CROSS,
cv::MorphShapes::MORPH_ELLIPSE),
Values(cv::compile_args(IMGPROC_FLUID))));
// GAPI/fluid does not support iterations parameter for the Dilate kernel
INSTANTIATE_TEST_CASE_P(DISABLED_Dilate3x3PerfTestFluid, Dilate3x3PerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(szVGA, sz720p, sz1080p),
Values(1, 2, 4),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid, SobelPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
Values(3), // add 5x5 once supported
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_16S, CV_32F),
@@ -23,8 +115,8 @@ namespace opencv_test
Values(1, 2),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid32F, SobelPerfTest,
Combine(Values(ToleranceFilter(1e-3f, 0.0).to_compare_f()),
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid32F, SobelPerfTest,
Combine(Values(ToleranceFilter(1e-3f, 0.0).to_compare_f()),
Values(CV_32FC1),
Values(3), // add 5x5 once supported
Values(szVGA, sz720p, sz1080p),
@@ -33,44 +125,44 @@ namespace opencv_test
Values(1, 2),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2GrayPerfTestFluid, RGB2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2GrayPerfTestFluid, RGB2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2GrayPerfTestFluid, BGR2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2GrayPerfTestFluid, BGR2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2YUVPerfTestFluid, RGB2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2YUVPerfTestFluid, RGB2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2RGBPerfTestFluid, YUV2RGBPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2RGBPerfTestFluid, YUV2RGBPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2YUVPerfTestFluid, BGR2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2YUVPerfTestFluid, BGR2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2BGRPerfTestFluid, YUV2BGRPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2BGRPerfTestFluid, YUV2BGRPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2LUVPerfTestFluid, BGR2LUVPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2LUVPerfTestFluid, BGR2LUVPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2LabPerfTestFluid, RGB2LabPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2LabPerfTestFluid, RGB2LabPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
}