2 #include "opencv2/imgproc.hpp"
3 #include "opencv2/highgui.hpp"
4 #include "opencv2/calib3d.hpp"
5 #include "opencv2/video.hpp"
6 #include "opencv2/cuda.hpp"
7 #include "opencv2/cudaimgproc.hpp"
8 #include "opencv2/cudaarithm.hpp"
9 #include "opencv2/cudawarping.hpp"
10 #include "opencv2/cudafeatures2d.hpp"
11 #include "opencv2/cudafilters.hpp"
12 #include "opencv2/cudaoptflow.hpp"
13 #include "opencv2/cudabgsegm.hpp"
15 #include "performance.h"
17 #include "opencv2/opencv_modules.hpp"
19 #ifdef HAVE_OPENCV_XFEATURES2D
20 #include "opencv2/xfeatures2d/cuda.hpp"
21 #include "opencv2/xfeatures2d/nonfree.hpp"
24 #ifdef HAVE_OPENCV_BGSEGM
25 #include "opencv2/bgsegm.hpp"
35 gen(src, 3000, 3000, CV_32F, 0, 1);
37 cuda::GpuMat d_src(src), d_templ, d_dst;
39 Ptr<cuda::TemplateMatching> alg = cuda::createTemplateMatching(src.type(), TM_CCORR);
41 for (int templ_size = 5; templ_size < 200; templ_size *= 5)
43 SUBTEST << src.cols << 'x' << src.rows << ", 32FC1" << ", templ " << templ_size << 'x' << templ_size << ", CCORR";
45 gen(templ, templ_size, templ_size, CV_32F, 0, 1);
46 matchTemplate(src, templ, dst, TM_CCORR);
49 matchTemplate(src, templ, dst, TM_CCORR);
52 d_templ.upload(templ);
53 alg->match(d_src, d_templ, d_dst);
56 alg->match(d_src, d_templ, d_dst);
67 double min_val, max_val;
68 Point min_loc, max_loc;
70 for (int size = 2000; size <= 8000; size *= 2)
72 SUBTEST << size << 'x' << size << ", 32F";
74 gen(src, size, size, CV_32F, 0, 1);
77 minMaxLoc(src, &min_val, &max_val, &min_loc, &max_loc);
83 cuda::minMaxLoc(d_src, &min_val, &max_val, &min_loc, &max_loc);
91 Mat src, dst, xmap, ymap;
92 cuda::GpuMat d_src, d_dst, d_xmap, d_ymap;
94 int interpolation = INTER_LINEAR;
95 int borderMode = BORDER_REPLICATE;
97 for (int size = 1000; size <= 4000; size *= 2)
99 SUBTEST << size << 'x' << size << ", 8UC4, INTER_LINEAR, BORDER_REPLICATE";
101 gen(src, size, size, CV_8UC4, 0, 256);
103 xmap.create(size, size, CV_32F);
104 ymap.create(size, size, CV_32F);
105 for (int i = 0; i < size; ++i)
107 float* xmap_row = xmap.ptr<float>(i);
108 float* ymap_row = ymap.ptr<float>(i);
109 for (int j = 0; j < size; ++j)
111 xmap_row[j] = (j - size * 0.5f) * 0.75f + size * 0.5f;
112 ymap_row[j] = (i - size * 0.5f) * 0.75f + size * 0.5f;
116 remap(src, dst, xmap, ymap, interpolation, borderMode);
119 remap(src, dst, xmap, ymap, interpolation, borderMode);
126 cuda::remap(d_src, d_dst, d_xmap, d_ymap, interpolation, borderMode);
129 cuda::remap(d_src, d_dst, d_xmap, d_ymap, interpolation, borderMode);
138 cuda::GpuMat d_src, d_dst;
140 for (int size = 1000; size <= 4000; size *= 2)
142 SUBTEST << size << 'x' << size << ", 32FC2, complex-to-complex";
144 gen(src, size, size, CV_32FC2, Scalar::all(0), Scalar::all(1));
154 cuda::dft(d_src, d_dst, Size(size, size));
157 cuda::dft(d_src, d_dst, Size(size, size));
166 cuda::GpuMat d_src, d_dst;
168 for (int size = 1000; size <= 4000; size *= 2)
170 SUBTEST << size << 'x' << size << ", 32FC1, BORDER_REFLECT101";
172 gen(src, size, size, CV_32F, 0, 1);
174 cornerHarris(src, dst, 5, 7, 0.1, BORDER_REFLECT101);
177 cornerHarris(src, dst, 5, 7, 0.1, BORDER_REFLECT101);
182 Ptr<cuda::CornernessCriteria> harris = cuda::createHarrisCorner(src.type(), 5, 7, 0.1, BORDER_REFLECT101);
184 harris->compute(d_src, d_dst);
187 harris->compute(d_src, d_dst);
196 cuda::GpuMat d_src, d_sum, d_buf;
198 for (int size = 1000; size <= 4000; size *= 2)
200 SUBTEST << size << 'x' << size << ", 8UC1";
202 gen(src, size, size, CV_8U, 0, 256);
212 cuda::integralBuffered(d_src, d_sum, d_buf);
215 cuda::integralBuffered(d_src, d_sum, d_buf);
224 cuda::GpuMat d_src, d_buf;
226 for (int size = 2000; size <= 4000; size += 1000)
228 SUBTEST << size << 'x' << size << ", 32FC4, NORM_INF";
230 gen(src, size, size, CV_32FC4, Scalar::all(0), Scalar::all(1));
240 cuda::norm(d_src, NORM_INF, d_buf);
243 cuda::norm(d_src, NORM_INF, d_buf);
251 int sp = 10, sr = 10;
254 cuda::GpuMat d_src, d_dst;
256 for (int size = 400; size <= 800; size *= 2)
258 SUBTEST << size << 'x' << size << ", 8UC3 vs 8UC4";
260 gen(src, size, size, CV_8UC3, Scalar::all(0), Scalar::all(256));
262 pyrMeanShiftFiltering(src, dst, sp, sr);
265 pyrMeanShiftFiltering(src, dst, sp, sr);
268 gen(src, size, size, CV_8UC4, Scalar::all(0), Scalar::all(256));
272 cuda::meanShiftFiltering(d_src, d_dst, sp, sr);
275 cuda::meanShiftFiltering(d_src, d_dst, sp, sr);
280 #ifdef HAVE_OPENCV_XFEATURES2D
284 Mat src = imread(abspath("aloeL.jpg"), IMREAD_GRAYSCALE);
285 if (src.empty()) throw runtime_error("can't open aloeL.jpg");
288 vector<KeyPoint> keypoints;
291 surf(src, Mat(), keypoints, descriptors);
294 surf(src, Mat(), keypoints, descriptors);
297 cuda::SURF_CUDA d_surf;
298 cuda::GpuMat d_src(src);
299 cuda::GpuMat d_keypoints;
300 cuda::GpuMat d_descriptors;
302 d_surf(d_src, cuda::GpuMat(), d_keypoints, d_descriptors);
305 d_surf(d_src, cuda::GpuMat(), d_keypoints, d_descriptors);
314 Mat src = imread(abspath("aloeL.jpg"), IMREAD_GRAYSCALE);
315 if (src.empty()) throw runtime_error("can't open aloeL.jpg");
317 vector<KeyPoint> keypoints;
319 FAST(src, keypoints, 20);
322 FAST(src, keypoints, 20);
325 cuda::FAST_CUDA d_FAST(20);
326 cuda::GpuMat d_src(src);
327 cuda::GpuMat d_keypoints;
329 d_FAST(d_src, cuda::GpuMat(), d_keypoints);
332 d_FAST(d_src, cuda::GpuMat(), d_keypoints);
339 Mat src = imread(abspath("aloeL.jpg"), IMREAD_GRAYSCALE);
340 if (src.empty()) throw runtime_error("can't open aloeL.jpg");
343 vector<KeyPoint> keypoints;
346 orb(src, Mat(), keypoints, descriptors);
349 orb(src, Mat(), keypoints, descriptors);
352 cuda::ORB_CUDA d_orb;
353 cuda::GpuMat d_src(src);
354 cuda::GpuMat d_keypoints;
355 cuda::GpuMat d_descriptors;
357 d_orb(d_src, cuda::GpuMat(), d_keypoints, d_descriptors);
360 d_orb(d_src, cuda::GpuMat(), d_keypoints, d_descriptors);
365 TEST(BruteForceMatcher)
371 BFMatcher matcher(NORM_L2);
374 gen(query, 3000, desc_len, CV_32F, 0, 1);
377 gen(train, 3000, desc_len, CV_32F, 0, 1);
381 cuda::BFMatcher_CUDA d_matcher(NORM_L2);
383 cuda::GpuMat d_query(query);
384 cuda::GpuMat d_train(train);
387 vector< vector<DMatch> > matches(2);
388 cuda::GpuMat d_trainIdx, d_distance, d_allDist, d_nMatches;
392 matcher.match(query, train, matches[0]);
395 matcher.match(query, train, matches[0]);
398 d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
401 d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
404 SUBTEST << "knnMatch";
406 matcher.knnMatch(query, train, matches, 2);
409 matcher.knnMatch(query, train, matches, 2);
412 d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
415 d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
418 SUBTEST << "radiusMatch";
420 float max_distance = 2.0f;
422 matcher.radiusMatch(query, train, matches, max_distance);
425 matcher.radiusMatch(query, train, matches, max_distance);
428 d_trainIdx.release();
430 d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, max_distance);
433 d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, max_distance);
441 cuda::GpuMat d_x, d_y, d_mag;
443 for (int size = 2000; size <= 4000; size += 1000)
445 SUBTEST << size << 'x' << size << ", 32FC1";
447 gen(x, size, size, CV_32F, 0, 1);
448 gen(y, size, size, CV_32F, 0, 1);
450 magnitude(x, y, mag);
453 magnitude(x, y, mag);
459 cuda::magnitude(d_x, d_y, d_mag);
462 cuda::magnitude(d_x, d_y, d_mag);
471 cuda::GpuMat d_src1, d_src2, d_dst;
473 for (int size = 2000; size <= 4000; size += 1000)
475 SUBTEST << size << 'x' << size << ", 32FC1";
477 gen(src1, size, size, CV_32F, 0, 1);
478 gen(src2, size, size, CV_32F, 0, 1);
480 add(src1, src2, dst);
483 add(src1, src2, dst);
489 cuda::add(d_src1, d_src2, d_dst);
492 cuda::add(d_src1, d_src2, d_dst);
501 cuda::GpuMat d_src, d_dst;
503 for (int size = 2000; size <= 4000; size += 1000)
505 SUBTEST << size << 'x' << size << ", 32F";
507 gen(src, size, size, CV_32F, 1, 10);
517 cuda::log(d_src, d_dst);
520 cuda::log(d_src, d_dst);
529 cuda::GpuMat d_src1, d_src2, d_dst;
531 for (int size = 2000; size <= 4000; size += 1000)
533 SUBTEST << size << 'x' << size;
535 gen(src1, size, size, CV_32FC2, Scalar::all(0), Scalar::all(1));
536 gen(src2, size, size, CV_32FC2, Scalar::all(0), Scalar::all(1));
538 mulSpectrums(src1, src2, dst, 0, true);
541 mulSpectrums(src1, src2, dst, 0, true);
547 cuda::mulSpectrums(d_src1, d_src2, d_dst, 0, true);
550 cuda::mulSpectrums(d_src1, d_src2, d_dst, 0, true);
559 cuda::GpuMat d_src, d_dst;
561 for (int size = 1000; size <= 3000; size += 1000)
563 SUBTEST << size << 'x' << size << ", 8UC4, up";
565 gen(src, size, size, CV_8UC4, 0, 256);
567 resize(src, dst, Size(), 2.0, 2.0);
570 resize(src, dst, Size(), 2.0, 2.0);
575 cuda::resize(d_src, d_dst, Size(), 2.0, 2.0);
578 cuda::resize(d_src, d_dst, Size(), 2.0, 2.0);
582 for (int size = 1000; size <= 3000; size += 1000)
584 SUBTEST << size << 'x' << size << ", 8UC4, down";
586 gen(src, size, size, CV_8UC4, 0, 256);
588 resize(src, dst, Size(), 0.5, 0.5);
591 resize(src, dst, Size(), 0.5, 0.5);
596 cuda::resize(d_src, d_dst, Size(), 0.5, 0.5);
599 cuda::resize(d_src, d_dst, Size(), 0.5, 0.5);
608 cuda::GpuMat d_src, d_dst;
610 gen(src, 4000, 4000, CV_8UC1, 0, 255);
613 SUBTEST << "4000x4000, 8UC1, COLOR_GRAY2BGRA";
615 cvtColor(src, dst, COLOR_GRAY2BGRA, 4);
618 cvtColor(src, dst, COLOR_GRAY2BGRA, 4);
621 cuda::cvtColor(d_src, d_dst, COLOR_GRAY2BGRA, 4);
624 cuda::cvtColor(d_src, d_dst, COLOR_GRAY2BGRA, 4);
630 SUBTEST << "4000x4000, 8UC3 vs 8UC4, COLOR_BGR2YCrCb";
632 cvtColor(src, dst, COLOR_BGR2YCrCb);
635 cvtColor(src, dst, COLOR_BGR2YCrCb);
638 cuda::cvtColor(d_src, d_dst, COLOR_BGR2YCrCb, 4);
641 cuda::cvtColor(d_src, d_dst, COLOR_BGR2YCrCb, 4);
647 SUBTEST << "4000x4000, 8UC4, COLOR_YCrCb2BGR";
649 cvtColor(src, dst, COLOR_YCrCb2BGR, 4);
652 cvtColor(src, dst, COLOR_YCrCb2BGR, 4);
655 cuda::cvtColor(d_src, d_dst, COLOR_YCrCb2BGR, 4);
658 cuda::cvtColor(d_src, d_dst, COLOR_YCrCb2BGR, 4);
664 SUBTEST << "4000x4000, 8UC3 vs 8UC4, COLOR_BGR2XYZ";
666 cvtColor(src, dst, COLOR_BGR2XYZ);
669 cvtColor(src, dst, COLOR_BGR2XYZ);
672 cuda::cvtColor(d_src, d_dst, COLOR_BGR2XYZ, 4);
675 cuda::cvtColor(d_src, d_dst, COLOR_BGR2XYZ, 4);
681 SUBTEST << "4000x4000, 8UC4, COLOR_XYZ2BGR";
683 cvtColor(src, dst, COLOR_XYZ2BGR, 4);
686 cvtColor(src, dst, COLOR_XYZ2BGR, 4);
689 cuda::cvtColor(d_src, d_dst, COLOR_XYZ2BGR, 4);
692 cuda::cvtColor(d_src, d_dst, COLOR_XYZ2BGR, 4);
698 SUBTEST << "4000x4000, 8UC3 vs 8UC4, COLOR_BGR2HSV";
700 cvtColor(src, dst, COLOR_BGR2HSV);
703 cvtColor(src, dst, COLOR_BGR2HSV);
706 cuda::cvtColor(d_src, d_dst, COLOR_BGR2HSV, 4);
709 cuda::cvtColor(d_src, d_dst, COLOR_BGR2HSV, 4);
715 SUBTEST << "4000x4000, 8UC4, COLOR_HSV2BGR";
717 cvtColor(src, dst, COLOR_HSV2BGR, 4);
720 cvtColor(src, dst, COLOR_HSV2BGR, 4);
723 cuda::cvtColor(d_src, d_dst, COLOR_HSV2BGR, 4);
726 cuda::cvtColor(d_src, d_dst, COLOR_HSV2BGR, 4);
737 cuda::GpuMat d_src, d_buf, d_dst;
739 for (int size = 2000; size <= 4000; size += 1000)
741 SUBTEST << size << 'x' << size;
743 gen(src, size, size, CV_8UC4, Scalar::all(0), Scalar::all(256));
744 ker = getStructuringElement(MORPH_RECT, Size(3, 3));
746 erode(src, dst, ker);
749 erode(src, dst, ker);
754 Ptr<cuda::Filter> erode = cuda::createMorphologyFilter(MORPH_ERODE, d_src.type(), ker);
756 erode->apply(d_src, d_dst);
759 erode->apply(d_src, d_dst);
767 cuda::GpuMat d_src, d_dst;
769 for (int size = 2000; size <= 4000; size += 1000)
771 SUBTEST << size << 'x' << size << ", 8UC1, THRESH_BINARY";
773 gen(src, size, size, CV_8U, 0, 100);
775 threshold(src, dst, 50.0, 0.0, THRESH_BINARY);
778 threshold(src, dst, 50.0, 0.0, THRESH_BINARY);
783 cuda::threshold(d_src, d_dst, 50.0, 0.0, THRESH_BINARY);
786 cuda::threshold(d_src, d_dst, 50.0, 0.0, THRESH_BINARY);
790 for (int size = 2000; size <= 4000; size += 1000)
792 SUBTEST << size << 'x' << size << ", 32FC1, THRESH_TRUNC [NPP]";
794 gen(src, size, size, CV_32FC1, 0, 100);
796 threshold(src, dst, 50.0, 0.0, THRESH_TRUNC);
799 threshold(src, dst, 50.0, 0.0, THRESH_TRUNC);
804 cuda::threshold(d_src, d_dst, 50.0, 0.0, THRESH_TRUNC);
807 cuda::threshold(d_src, d_dst, 50.0, 0.0, THRESH_TRUNC);
815 cuda::GpuMat d_src, d_dst;
817 for (int size = 1000; size <= 4000; size += 1000)
819 SUBTEST << size << 'x' << size << ", 32F";
821 gen(src, size, size, CV_32F, 0, 100);
831 cuda::pow(d_src, -2.0, d_dst);
834 cuda::pow(d_src, -2.0, d_dst);
844 cuda::GpuMat d_src, d_dst;
846 Mat rvec; gen(rvec, 1, 3, CV_32F, 0, 1);
847 Mat tvec; gen(tvec, 1, 3, CV_32F, 0, 1);
848 Mat camera_mat; gen(camera_mat, 3, 3, CV_32F, 0, 1);
849 camera_mat.at<float>(0, 1) = 0.f;
850 camera_mat.at<float>(1, 0) = 0.f;
851 camera_mat.at<float>(2, 0) = 0.f;
852 camera_mat.at<float>(2, 1) = 0.f;
854 for (int size = (int)1e6, count = 0; size >= 1e5 && count < 5; size = int(size / 1.4), count++)
858 gen(src, 1, size, CV_32FC3, Scalar::all(0), Scalar::all(10));
860 projectPoints(src, rvec, tvec, camera_mat, Mat::zeros(1, 8, CV_32F), dst);
863 projectPoints(src, rvec, tvec, camera_mat, Mat::zeros(1, 8, CV_32F), dst);
868 cuda::projectPoints(d_src, rvec, tvec, camera_mat, Mat(), d_dst);
871 cuda::projectPoints(d_src, rvec, tvec, camera_mat, Mat(), d_dst);
877 static void InitSolvePnpRansac()
879 Mat object; gen(object, 1, 4, CV_32FC3, Scalar::all(0), Scalar::all(100));
880 Mat image; gen(image, 1, 4, CV_32FC2, Scalar::all(0), Scalar::all(100));
882 cuda::solvePnPRansac(object, image, Mat::eye(3, 3, CV_32F), Mat(), rvec, tvec);
888 InitSolvePnpRansac();
890 for (int num_points = 5000; num_points <= 300000; num_points = int(num_points * 3.76))
892 SUBTEST << num_points;
894 Mat object; gen(object, 1, num_points, CV_32FC3, Scalar::all(10), Scalar::all(100));
895 Mat image; gen(image, 1, num_points, CV_32FC2, Scalar::all(10), Scalar::all(100));
896 Mat camera_mat; gen(camera_mat, 3, 3, CV_32F, 0.5, 1);
897 camera_mat.at<float>(0, 1) = 0.f;
898 camera_mat.at<float>(1, 0) = 0.f;
899 camera_mat.at<float>(2, 0) = 0.f;
900 camera_mat.at<float>(2, 1) = 0.f;
903 const int num_iters = 200;
904 const float max_dist = 2.0f;
905 vector<int> inliers_cpu, inliers_gpu;
908 solvePnPRansac(object, image, camera_mat, Mat::zeros(1, 8, CV_32F), rvec, tvec, false, num_iters,
909 max_dist, int(num_points * 0.05), inliers_cpu);
913 cuda::solvePnPRansac(object, image, camera_mat, Mat::zeros(1, 8, CV_32F), rvec, tvec, false, num_iters,
914 max_dist, int(num_points * 0.05), &inliers_gpu);
921 for (int size = 1000; size <= 4000; size += 1000)
923 SUBTEST << size << 'x' << size << ", 8UC4";
927 gen(src, size, size, CV_8UC4, 0, 256);
929 GaussianBlur(src, dst, Size(3, 3), 1);
932 GaussianBlur(src, dst, Size(3, 3), 1);
935 cuda::GpuMat d_src(src);
936 cuda::GpuMat d_dst(src.size(), src.type());
939 cv::Ptr<cv::cuda::Filter> gauss = cv::cuda::createGaussianFilter(d_src.type(), -1, cv::Size(3, 3), 1);
941 gauss->apply(d_src, d_dst);
944 gauss->apply(d_src, d_dst);
951 for (int size = 512; size <= 2048; size *= 2)
954 gen(src, size, size, CV_8UC4, 0, 256);
956 for (int ksize = 3; ksize <= 16; ksize += 2)
958 SUBTEST << "ksize = " << ksize << ", " << size << 'x' << size << ", 8UC4";
961 gen(kernel, ksize, ksize, CV_32FC1, 0.0, 1.0);
964 cv::filter2D(src, dst, -1, kernel);
967 cv::filter2D(src, dst, -1, kernel);
970 cuda::GpuMat d_src(src);
973 Ptr<cuda::Filter> filter2D = cuda::createLinearFilter(d_src.type(), -1, kernel);
974 filter2D->apply(d_src, d_dst);
977 filter2D->apply(d_src, d_dst);
985 for (int size = 4000; size >= 1000; size -= 1000)
987 SUBTEST << size << 'x' << size << ", 8UC4";
990 gen(src, size, size, CV_8UC4, 0, 256);
998 cuda::GpuMat d_src(src);
1001 cuda::pyrDown(d_src, d_dst);
1004 cuda::pyrDown(d_src, d_dst);
1011 for (int size = 2000; size >= 1000; size -= 1000)
1013 SUBTEST << size << 'x' << size << ", 8UC4";
1017 gen(src, size, size, CV_8UC4, 0, 256);
1025 cuda::GpuMat d_src(src);
1028 cuda::pyrUp(d_src, d_dst);
1031 cuda::pyrUp(d_src, d_dst);
1039 for (int size = 1000; size < 4000; size += 1000)
1041 SUBTEST << size << 'x' << size;
1045 gen(src, size, size, CV_8UC1, 0, 256);
1047 equalizeHist(src, dst);
1050 equalizeHist(src, dst);
1053 cuda::GpuMat d_src(src);
1057 cuda::equalizeHist(d_src, d_dst, d_buf);
1060 cuda::equalizeHist(d_src, d_dst, d_buf);
1068 Mat img = imread(abspath("aloeL.jpg"), IMREAD_GRAYSCALE);
1070 if (img.empty()) throw runtime_error("can't open aloeL.jpg");
1072 Mat edges(img.size(), CV_8UC1);
1075 Canny(img, edges, 50.0, 100.0);
1078 cuda::GpuMat d_img(img);
1079 cuda::GpuMat d_edges;
1081 Ptr<cuda::CannyEdgeDetector> canny = cuda::createCannyEdgeDetector(50.0, 100.0);
1083 canny->detect(d_img, d_edges);
1086 canny->detect(d_img, d_edges);
1093 for (int size = 1000; size < 4000; size += 1000)
1096 gen(src, size, size, CV_32F, 0, 255);
1101 cuda::GpuMat d_src(src);
1102 cuda::GpuMat d_dst0;
1103 cuda::GpuMat d_dst1;
1105 SUBTEST << size << 'x' << size << ", dim = 0";
1107 reduce(src, dst0, 0, REDUCE_MIN);
1110 reduce(src, dst0, 0, REDUCE_MIN);
1113 cuda::reduce(d_src, d_dst0, 0, REDUCE_MIN);
1116 cuda::reduce(d_src, d_dst0, 0, REDUCE_MIN);
1119 SUBTEST << size << 'x' << size << ", dim = 1";
1121 reduce(src, dst1, 1, REDUCE_MIN);
1124 reduce(src, dst1, 1, REDUCE_MIN);
1127 cuda::reduce(d_src, d_dst1, 1, REDUCE_MIN);
1130 cuda::reduce(d_src, d_dst1, 1, REDUCE_MIN);
1138 Mat src1, src2, src3, dst;
1139 cuda::GpuMat d_src1, d_src2, d_src3, d_dst;
1141 for (int size = 512; size <= 1024; size *= 2)
1143 SUBTEST << size << 'x' << size;
1145 gen(src1, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
1146 gen(src2, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
1147 gen(src3, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
1149 gemm(src1, src2, 1.0, src3, 1.0, dst);
1152 gemm(src1, src2, 1.0, src3, 1.0, dst);
1155 d_src1.upload(src1);
1156 d_src2.upload(src2);
1157 d_src3.upload(src3);
1159 cuda::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
1162 cuda::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
1167 TEST(GoodFeaturesToTrack)
1169 Mat src = imread(abspath("aloeL.jpg"), IMREAD_GRAYSCALE);
1170 if (src.empty()) throw runtime_error("can't open aloeL.jpg");
1172 vector<Point2f> pts;
1174 goodFeaturesToTrack(src, pts, 8000, 0.01, 0.0);
1177 goodFeaturesToTrack(src, pts, 8000, 0.01, 0.0);
1180 Ptr<cuda::CornersDetector> detector = cuda::createGoodFeaturesToTrackDetector(src.type(), 8000, 0.01, 0.0);
1182 cuda::GpuMat d_src(src);
1185 detector->detect(d_src, d_pts);
1188 detector->detect(d_src, d_pts);
1192 TEST(PyrLKOpticalFlow)
1194 Mat frame0 = imread(abspath("rubberwhale1.png"));
1195 if (frame0.empty()) throw runtime_error("can't open rubberwhale1.png");
1197 Mat frame1 = imread(abspath("rubberwhale2.png"));
1198 if (frame1.empty()) throw runtime_error("can't open rubberwhale2.png");
1201 cvtColor(frame0, gray_frame, COLOR_BGR2GRAY);
1203 for (int points = 1000; points <= 8000; points *= 2)
1207 vector<Point2f> pts;
1208 goodFeaturesToTrack(gray_frame, pts, points, 0.01, 0.0);
1210 vector<Point2f> nextPts;
1211 vector<unsigned char> status;
1215 calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
1218 calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
1221 cuda::PyrLKOpticalFlow d_pyrLK;
1223 cuda::GpuMat d_frame0(frame0);
1224 cuda::GpuMat d_frame1(frame1);
1227 Mat pts_mat(1, (int)pts.size(), CV_32FC2, (void*)&pts[0]);
1228 d_pts.upload(pts_mat);
1230 cuda::GpuMat d_nextPts;
1231 cuda::GpuMat d_status;
1234 d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
1237 d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
1243 TEST(FarnebackOpticalFlow)
1245 const string datasets[] = {"rubberwhale", "basketball"};
1246 for (size_t i = 0; i < sizeof(datasets)/sizeof(*datasets); ++i) {
1247 for (int fastPyramids = 0; fastPyramids < 2; ++fastPyramids) {
1248 for (int useGaussianBlur = 0; useGaussianBlur < 2; ++useGaussianBlur) {
1250 SUBTEST << "dataset=" << datasets[i] << ", fastPyramids=" << fastPyramids << ", useGaussianBlur=" << useGaussianBlur;
1251 Mat frame0 = imread(abspath(datasets[i] + "1.png"), IMREAD_GRAYSCALE);
1252 Mat frame1 = imread(abspath(datasets[i] + "2.png"), IMREAD_GRAYSCALE);
1253 if (frame0.empty()) throw runtime_error("can't open " + datasets[i] + "1.png");
1254 if (frame1.empty()) throw runtime_error("can't open " + datasets[i] + "2.png");
1256 cuda::FarnebackOpticalFlow calc;
1257 calc.fastPyramids = fastPyramids != 0;
1258 calc.flags |= useGaussianBlur ? OPTFLOW_FARNEBACK_GAUSSIAN : 0;
1260 cuda::GpuMat d_frame0(frame0), d_frame1(frame1), d_flowx, d_flowy;
1262 calc(d_frame0, d_frame1, d_flowx, d_flowy);
1267 calcOpticalFlowFarneback(frame0, frame1, flow, calc.pyrScale, calc.numLevels, calc.winSize, calc.numIters, calc.polyN, calc.polySigma, calc.flags);
1273 #ifdef HAVE_OPENCV_BGSEGM
1277 const std::string inputFile = abspath("768x576.avi");
1279 cv::VideoCapture cap(inputFile);
1280 if (!cap.isOpened()) throw runtime_error("can't open 768x576.avi");
1285 cv::Ptr<cv::BackgroundSubtractor> mog = cv::bgsegm::createBackgroundSubtractorMOG();
1288 mog->apply(frame, foreground, 0.01);
1290 while (!TestSystem::instance().stop())
1294 TestSystem::instance().cpuOn();
1296 mog->apply(frame, foreground, 0.01);
1298 TestSystem::instance().cpuOff();
1300 TestSystem::instance().cpuComplete();
1302 cap.open(inputFile);
1306 cv::cuda::GpuMat d_frame(frame);
1307 cv::Ptr<cv::BackgroundSubtractor> d_mog = cv::cuda::createBackgroundSubtractorMOG();
1308 cv::cuda::GpuMat d_foreground;
1310 d_mog->apply(d_frame, d_foreground, 0.01);
1312 while (!TestSystem::instance().stop())
1315 d_frame.upload(frame);
1317 TestSystem::instance().gpuOn();
1319 d_mog->apply(d_frame, d_foreground, 0.01);
1321 TestSystem::instance().gpuOff();
1323 TestSystem::instance().gpuComplete();
1330 const std::string inputFile = abspath("768x576.avi");
1332 cv::VideoCapture cap(inputFile);
1333 if (!cap.isOpened()) throw runtime_error("can't open 768x576.avi");
1338 cv::Ptr<cv::BackgroundSubtractor> mog2 = cv::createBackgroundSubtractorMOG2();
1342 mog2->apply(frame, foreground);
1343 mog2->getBackgroundImage(background);
1345 while (!TestSystem::instance().stop())
1349 TestSystem::instance().cpuOn();
1351 mog2->apply(frame, foreground);
1352 mog2->getBackgroundImage(background);
1354 TestSystem::instance().cpuOff();
1356 TestSystem::instance().cpuComplete();
1358 cap.open(inputFile);
1362 cv::Ptr<cv::BackgroundSubtractor> d_mog2 = cv::cuda::createBackgroundSubtractorMOG2();
1363 cv::cuda::GpuMat d_frame(frame);
1364 cv::cuda::GpuMat d_foreground;
1365 cv::cuda::GpuMat d_background;
1367 d_mog2->apply(d_frame, d_foreground);
1368 d_mog2->getBackgroundImage(d_background);
1370 while (!TestSystem::instance().stop())
1373 d_frame.upload(frame);
1375 TestSystem::instance().gpuOn();
1377 d_mog2->apply(d_frame, d_foreground);
1378 d_mog2->getBackgroundImage(d_background);
1380 TestSystem::instance().gpuOff();
1382 TestSystem::instance().gpuComplete();