1 /*M///////////////////////////////////////////////////////////////////////////////////////
\r
3 // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
\r
5 // By downloading, copying, installing or using the software you agree to this license.
\r
6 // If you do not agree to this license, do not download, install,
\r
7 // copy or use the software.
\r
10 // Intel License Agreement
\r
11 // For Open Source Computer Vision Library
\r
13 // Copyright (C) 2000, Intel Corporation, all rights reserved.
\r
14 // Third party copyrights are property of their respective owners.
\r
16 // Redistribution and use in source and binary forms, with or without modification,
\r
17 // are permitted provided that the following conditions are met:
\r
19 // * Redistribution's of source code must retain the above copyright notice,
\r
20 // this list of conditions and the following disclaimer.
\r
22 // * Redistribution's in binary form must reproduce the above copyright notice,
\r
23 // this list of conditions and the following disclaimer in the documentation
\r
24 // and/or other materials provided with the distribution.
\r
26 // * The name of Intel Corporation may not be used to endorse or promote products
\r
27 // derived from this software without specific prior written permission.
\r
29 // This software is provided by the copyright holders and contributors "as is" and
\r
30 // any express or implied warranties, including, but not limited to, the implied
\r
31 // warranties of merchantability and fitness for a particular purpose are disclaimed.
\r
32 // In no event shall the Intel Corporation or contributors be liable for any direct,
\r
33 // indirect, incidental, special, exemplary, or consequential damages
\r
34 // (including, but not limited to, procurement of substitute goods or services;
\r
35 // loss of use, data, or profits; or business interruption) however caused
\r
36 // and on any theory of liability, whether in contract, strict liability,
\r
37 // or tort (including negligence or otherwise) arising in any way out of
\r
38 // the use of this software, even if advised of the possibility of such damage.
\r
42 #include "test_precomp.hpp"
\r
48 ///////////////////////////////////////////////////////////////////////////////////////////////////////
\r
51 PARAM_TEST_CASE(Integral, cv::gpu::DeviceInfo, cv::Size, UseRoi)
\r
53 cv::gpu::DeviceInfo devInfo;
\r
57 virtual void SetUp()
\r
59 devInfo = GET_PARAM(0);
\r
60 size = GET_PARAM(1);
\r
61 useRoi = GET_PARAM(2);
\r
63 cv::gpu::setDevice(devInfo.deviceID());
\r
67 TEST_P(Integral, Accuracy)
\r
69 cv::Mat src = randomMat(size, CV_8UC1);
\r
71 cv::gpu::GpuMat dst = createMat(cv::Size(src.cols + 1, src.rows + 1), CV_32SC1, useRoi);
\r
72 cv::gpu::integral(loadMat(src, useRoi), dst);
\r
75 cv::integral(src, dst_gold, CV_32S);
\r
77 EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
\r
80 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Integral, testing::Combine(
\r
85 ///////////////////////////////////////////////////////////////////////////////////////////////////////
\r
88 struct HistEven : testing::TestWithParam<cv::gpu::DeviceInfo>
\r
90 cv::gpu::DeviceInfo devInfo;
\r
92 virtual void SetUp()
\r
94 devInfo = GetParam();
\r
96 cv::gpu::setDevice(devInfo.deviceID());
\r
100 TEST_P(HistEven, Accuracy)
\r
102 cv::Mat img = readImage("stereobm/aloe-L.png");
\r
103 ASSERT_FALSE(img.empty());
\r
106 cv::cvtColor(img, hsv, CV_BGR2HSV);
\r
109 float hranges[] = {0.0f, 180.0f};
\r
111 std::vector<cv::gpu::GpuMat> srcs;
\r
112 cv::gpu::split(loadMat(hsv), srcs);
\r
114 cv::gpu::GpuMat hist;
\r
115 cv::gpu::histEven(srcs[0], hist, hbins, (int)hranges[0], (int)hranges[1]);
\r
118 int histSize[] = {hbins};
\r
119 const float* ranges[] = {hranges};
\r
120 int channels[] = {0};
\r
121 cv::calcHist(&hsv, 1, channels, cv::Mat(), histnd, 1, histSize, ranges);
\r
123 cv::Mat hist_gold = histnd;
\r
124 hist_gold = hist_gold.t();
\r
125 hist_gold.convertTo(hist_gold, CV_32S);
\r
127 EXPECT_MAT_NEAR(hist_gold, hist, 0.0);
\r
130 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, HistEven, ALL_DEVICES);
\r
132 ///////////////////////////////////////////////////////////////////////////////////////////////////////
\r
135 void calcHistGold(const cv::Mat& src, cv::Mat& hist)
\r
137 hist.create(1, 256, CV_32SC1);
\r
138 hist.setTo(cv::Scalar::all(0));
\r
140 int* hist_row = hist.ptr<int>();
\r
141 for (int y = 0; y < src.rows; ++y)
\r
143 const uchar* src_row = src.ptr(y);
\r
145 for (int x = 0; x < src.cols; ++x)
\r
146 ++hist_row[src_row[x]];
\r
150 PARAM_TEST_CASE(CalcHist, cv::gpu::DeviceInfo, cv::Size)
\r
152 cv::gpu::DeviceInfo devInfo;
\r
156 virtual void SetUp()
\r
158 devInfo = GET_PARAM(0);
\r
159 size = GET_PARAM(1);
\r
161 cv::gpu::setDevice(devInfo.deviceID());
\r
165 TEST_P(CalcHist, Accuracy)
\r
167 cv::Mat src = randomMat(size, CV_8UC1);
\r
169 cv::gpu::GpuMat hist;
\r
170 cv::gpu::calcHist(loadMat(src), hist);
\r
173 calcHistGold(src, hist_gold);
\r
175 EXPECT_MAT_NEAR(hist_gold, hist, 0.0);
\r
178 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CalcHist, testing::Combine(
\r
182 ///////////////////////////////////////////////////////////////////////////////////////////////////////
\r
185 PARAM_TEST_CASE(EqualizeHist, cv::gpu::DeviceInfo, cv::Size)
\r
187 cv::gpu::DeviceInfo devInfo;
\r
190 virtual void SetUp()
\r
192 devInfo = GET_PARAM(0);
\r
193 size = GET_PARAM(1);
\r
195 cv::gpu::setDevice(devInfo.deviceID());
\r
199 TEST_P(EqualizeHist, Accuracy)
\r
201 cv::Mat src = randomMat(size, CV_8UC1);
\r
203 cv::gpu::GpuMat dst;
\r
204 cv::gpu::equalizeHist(loadMat(src), dst);
\r
207 cv::equalizeHist(src, dst_gold);
\r
209 EXPECT_MAT_NEAR(dst_gold, dst, 3.0);
\r
212 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, EqualizeHist, testing::Combine(
\r
216 ////////////////////////////////////////////////////////////////////////
\r
219 PARAM_TEST_CASE(ColumnSum, cv::gpu::DeviceInfo, cv::Size)
\r
221 cv::gpu::DeviceInfo devInfo;
\r
224 virtual void SetUp()
\r
226 devInfo = GET_PARAM(0);
\r
227 size = GET_PARAM(1);
\r
229 cv::gpu::setDevice(devInfo.deviceID());
\r
233 TEST_P(ColumnSum, Accuracy)
\r
235 cv::Mat src = randomMat(size, CV_32FC1);
\r
237 cv::gpu::GpuMat d_dst;
\r
238 cv::gpu::columnSum(loadMat(src), d_dst);
\r
240 cv::Mat dst(d_dst);
\r
242 for (int j = 0; j < src.cols; ++j)
\r
244 float gold = src.at<float>(0, j);
\r
245 float res = dst.at<float>(0, j);
\r
246 ASSERT_NEAR(res, gold, 1e-5);
\r
249 for (int i = 1; i < src.rows; ++i)
\r
251 for (int j = 0; j < src.cols; ++j)
\r
253 float gold = src.at<float>(i, j) += src.at<float>(i - 1, j);
\r
254 float res = dst.at<float>(i, j);
\r
255 ASSERT_NEAR(res, gold, 1e-5);
\r
260 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, ColumnSum, testing::Combine(
\r
264 ////////////////////////////////////////////////////////
\r
267 IMPLEMENT_PARAM_CLASS(AppertureSize, int);
\r
268 IMPLEMENT_PARAM_CLASS(L2gradient, bool);
\r
270 PARAM_TEST_CASE(Canny, cv::gpu::DeviceInfo, AppertureSize, L2gradient, UseRoi)
\r
272 cv::gpu::DeviceInfo devInfo;
\r
273 int apperture_size;
\r
274 bool useL2gradient;
\r
277 virtual void SetUp()
\r
279 devInfo = GET_PARAM(0);
\r
280 apperture_size = GET_PARAM(1);
\r
281 useL2gradient = GET_PARAM(2);
\r
282 useRoi = GET_PARAM(3);
\r
284 cv::gpu::setDevice(devInfo.deviceID());
\r
288 TEST_P(Canny, Accuracy)
\r
290 cv::Mat img = readImage("stereobm/aloe-L.png", cv::IMREAD_GRAYSCALE);
\r
291 ASSERT_FALSE(img.empty());
\r
293 double low_thresh = 50.0;
\r
294 double high_thresh = 100.0;
\r
296 if (!supportFeature(devInfo, cv::gpu::SHARED_ATOMICS))
\r
300 cv::gpu::GpuMat edges;
\r
301 cv::gpu::Canny(loadMat(img), edges, low_thresh, high_thresh, apperture_size, useL2gradient);
\r
303 catch (const cv::Exception& e)
\r
305 ASSERT_EQ(CV_StsNotImplemented, e.code);
\r
310 cv::gpu::GpuMat edges;
\r
311 cv::gpu::Canny(loadMat(img, useRoi), edges, low_thresh, high_thresh, apperture_size, useL2gradient);
\r
313 cv::Mat edges_gold;
\r
314 cv::Canny(img, edges_gold, low_thresh, high_thresh, apperture_size, useL2gradient);
\r
316 EXPECT_MAT_SIMILAR(edges_gold, edges, 1e-2);
\r
320 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Canny, testing::Combine(
\r
322 testing::Values(AppertureSize(3), AppertureSize(5)),
\r
323 testing::Values(L2gradient(false), L2gradient(true)),
\r
326 ////////////////////////////////////////////////////////////////////////////////
\r
329 struct MeanShift : testing::TestWithParam<cv::gpu::DeviceInfo>
\r
331 cv::gpu::DeviceInfo devInfo;
\r
338 virtual void SetUp()
\r
340 devInfo = GetParam();
\r
342 cv::gpu::setDevice(devInfo.deviceID());
\r
344 img = readImageType("meanshift/cones.png", CV_8UC4);
\r
345 ASSERT_FALSE(img.empty());
\r
352 TEST_P(MeanShift, Filtering)
\r
354 cv::Mat img_template;
\r
355 if (supportFeature(devInfo, cv::gpu::FEATURE_SET_COMPUTE_20))
\r
356 img_template = readImage("meanshift/con_result.png");
\r
358 img_template = readImage("meanshift/con_result_CC1X.png");
\r
359 ASSERT_FALSE(img_template.empty());
\r
361 cv::gpu::GpuMat d_dst;
\r
362 cv::gpu::meanShiftFiltering(loadMat(img), d_dst, spatialRad, colorRad);
\r
364 ASSERT_EQ(CV_8UC4, d_dst.type());
\r
366 cv::Mat dst(d_dst);
\r
369 cv::cvtColor(dst, result, CV_BGRA2BGR);
\r
371 EXPECT_MAT_NEAR(img_template, result, 0.0);
\r
374 TEST_P(MeanShift, Proc)
\r
376 cv::FileStorage fs;
\r
377 if (supportFeature(devInfo, cv::gpu::FEATURE_SET_COMPUTE_20))
\r
378 fs.open(std::string(cvtest::TS::ptr()->get_data_path()) + "meanshift/spmap.yaml", cv::FileStorage::READ);
\r
380 fs.open(std::string(cvtest::TS::ptr()->get_data_path()) + "meanshift/spmap_CC1X.yaml", cv::FileStorage::READ);
\r
381 ASSERT_TRUE(fs.isOpened());
\r
383 cv::Mat spmap_template;
\r
384 fs["spmap"] >> spmap_template;
\r
385 ASSERT_FALSE(spmap_template.empty());
\r
387 cv::gpu::GpuMat rmap_filtered;
\r
388 cv::gpu::meanShiftFiltering(loadMat(img), rmap_filtered, spatialRad, colorRad);
\r
390 cv::gpu::GpuMat rmap;
\r
391 cv::gpu::GpuMat spmap;
\r
392 cv::gpu::meanShiftProc(loadMat(img), rmap, spmap, spatialRad, colorRad);
\r
394 ASSERT_EQ(CV_8UC4, rmap.type());
\r
396 EXPECT_MAT_NEAR(rmap_filtered, rmap, 0.0);
\r
397 EXPECT_MAT_NEAR(spmap_template, spmap, 0.0);
\r
400 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MeanShift, ALL_DEVICES);
\r
402 ////////////////////////////////////////////////////////////////////////////////
\r
403 // MeanShiftSegmentation
\r
405 IMPLEMENT_PARAM_CLASS(MinSize, int);
\r
407 PARAM_TEST_CASE(MeanShiftSegmentation, cv::gpu::DeviceInfo, MinSize)
\r
409 cv::gpu::DeviceInfo devInfo;
\r
412 virtual void SetUp()
\r
414 devInfo = GET_PARAM(0);
\r
415 minsize = GET_PARAM(1);
\r
417 cv::gpu::setDevice(devInfo.deviceID());
\r
421 TEST_P(MeanShiftSegmentation, Regression)
\r
423 cv::Mat img = readImageType("meanshift/cones.png", CV_8UC4);
\r
424 ASSERT_FALSE(img.empty());
\r
426 std::ostringstream path;
\r
427 path << "meanshift/cones_segmented_sp10_sr10_minsize" << minsize;
\r
428 if (supportFeature(devInfo, cv::gpu::FEATURE_SET_COMPUTE_20))
\r
431 path << "_CC1X.png";
\r
432 cv::Mat dst_gold = readImage(path.str());
\r
433 ASSERT_FALSE(dst_gold.empty());
\r
436 cv::gpu::meanShiftSegmentation(loadMat(img), dst, 10, 10, minsize);
\r
439 cv::cvtColor(dst, dst_rgb, CV_BGRA2BGR);
\r
441 EXPECT_MAT_SIMILAR(dst_gold, dst_rgb, 1e-3);
\r
444 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MeanShiftSegmentation, testing::Combine(
\r
446 testing::Values(MinSize(0), MinSize(4), MinSize(20), MinSize(84), MinSize(340), MinSize(1364))));
\r
448 ////////////////////////////////////////////////////////////////////////////
\r
451 template <typename T>
\r
452 void blendLinearGold(const cv::Mat& img1, const cv::Mat& img2, const cv::Mat& weights1, const cv::Mat& weights2, cv::Mat& result_gold)
\r
454 result_gold.create(img1.size(), img1.type());
\r
456 int cn = img1.channels();
\r
458 for (int y = 0; y < img1.rows; ++y)
\r
460 const float* weights1_row = weights1.ptr<float>(y);
\r
461 const float* weights2_row = weights2.ptr<float>(y);
\r
462 const T* img1_row = img1.ptr<T>(y);
\r
463 const T* img2_row = img2.ptr<T>(y);
\r
464 T* result_gold_row = result_gold.ptr<T>(y);
\r
466 for (int x = 0; x < img1.cols * cn; ++x)
\r
468 float w1 = weights1_row[x / cn];
\r
469 float w2 = weights2_row[x / cn];
\r
470 result_gold_row[x] = static_cast<T>((img1_row[x] * w1 + img2_row[x] * w2) / (w1 + w2 + 1e-5f));
\r
475 PARAM_TEST_CASE(Blend, cv::gpu::DeviceInfo, cv::Size, MatType, UseRoi)
\r
477 cv::gpu::DeviceInfo devInfo;
\r
482 virtual void SetUp()
\r
484 devInfo = GET_PARAM(0);
\r
485 size = GET_PARAM(1);
\r
486 type = GET_PARAM(2);
\r
487 useRoi = GET_PARAM(3);
\r
489 cv::gpu::setDevice(devInfo.deviceID());
\r
493 TEST_P(Blend, Accuracy)
\r
495 int depth = CV_MAT_DEPTH(type);
\r
497 cv::Mat img1 = randomMat(size, type, 0.0, depth == CV_8U ? 255.0 : 1.0);
\r
498 cv::Mat img2 = randomMat(size, type, 0.0, depth == CV_8U ? 255.0 : 1.0);
\r
499 cv::Mat weights1 = randomMat(size, CV_32F, 0, 1);
\r
500 cv::Mat weights2 = randomMat(size, CV_32F, 0, 1);
\r
502 cv::gpu::GpuMat result;
\r
503 cv::gpu::blendLinear(loadMat(img1, useRoi), loadMat(img2, useRoi), loadMat(weights1, useRoi), loadMat(weights2, useRoi), result);
\r
505 cv::Mat result_gold;
\r
506 if (depth == CV_8U)
\r
507 blendLinearGold<uchar>(img1, img2, weights1, weights2, result_gold);
\r
509 blendLinearGold<float>(img1, img2, weights1, weights2, result_gold);
\r
511 EXPECT_MAT_NEAR(result_gold, result, CV_MAT_DEPTH(type) == CV_8U ? 1.0 : 1e-5);
\r
514 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Blend, testing::Combine(
\r
517 testing::Values(MatType(CV_8UC1), MatType(CV_8UC3), MatType(CV_8UC4), MatType(CV_32FC1), MatType(CV_32FC3), MatType(CV_32FC4)),
\r
520 ////////////////////////////////////////////////////////
\r
523 void convolveDFT(const cv::Mat& A, const cv::Mat& B, cv::Mat& C, bool ccorr = false)
\r
525 // reallocate the output array if needed
\r
526 C.create(std::abs(A.rows - B.rows) + 1, std::abs(A.cols - B.cols) + 1, A.type());
\r
529 // compute the size of DFT transform
\r
530 dftSize.width = cv::getOptimalDFTSize(A.cols + B.cols - 1);
\r
531 dftSize.height = cv::getOptimalDFTSize(A.rows + B.rows - 1);
\r
533 // allocate temporary buffers and initialize them with 0s
\r
534 cv::Mat tempA(dftSize, A.type(), cv::Scalar::all(0));
\r
535 cv::Mat tempB(dftSize, B.type(), cv::Scalar::all(0));
\r
537 // copy A and B to the top-left corners of tempA and tempB, respectively
\r
538 cv::Mat roiA(tempA, cv::Rect(0, 0, A.cols, A.rows));
\r
540 cv::Mat roiB(tempB, cv::Rect(0, 0, B.cols, B.rows));
\r
543 // now transform the padded A & B in-place;
\r
544 // use "nonzeroRows" hint for faster processing
\r
545 cv::dft(tempA, tempA, 0, A.rows);
\r
546 cv::dft(tempB, tempB, 0, B.rows);
\r
548 // multiply the spectrums;
\r
549 // the function handles packed spectrum representations well
\r
550 cv::mulSpectrums(tempA, tempB, tempA, 0, ccorr);
\r
552 // transform the product back from the frequency domain.
\r
553 // Even though all the result rows will be non-zero,
\r
554 // you need only the first C.rows of them, and thus you
\r
555 // pass nonzeroRows == C.rows
\r
556 cv::dft(tempA, tempA, cv::DFT_INVERSE + cv::DFT_SCALE, C.rows);
\r
558 // now copy the result back to C.
\r
559 tempA(cv::Rect(0, 0, C.cols, C.rows)).copyTo(C);
\r
562 IMPLEMENT_PARAM_CLASS(KSize, int);
\r
563 IMPLEMENT_PARAM_CLASS(Ccorr, bool);
\r
565 PARAM_TEST_CASE(Convolve, cv::gpu::DeviceInfo, cv::Size, KSize, Ccorr)
\r
567 cv::gpu::DeviceInfo devInfo;
\r
572 virtual void SetUp()
\r
574 devInfo = GET_PARAM(0);
\r
575 size = GET_PARAM(1);
\r
576 ksize = GET_PARAM(2);
\r
577 ccorr = GET_PARAM(3);
\r
579 cv::gpu::setDevice(devInfo.deviceID());
\r
583 TEST_P(Convolve, Accuracy)
\r
585 cv::Mat src = randomMat(size, CV_32FC1, 0.0, 100.0);
\r
586 cv::Mat kernel = randomMat(cv::Size(ksize, ksize), CV_32FC1, 0.0, 1.0);
\r
588 cv::gpu::GpuMat dst;
\r
589 cv::gpu::convolve(loadMat(src), loadMat(kernel), dst, ccorr);
\r
592 convolveDFT(src, kernel, dst_gold, ccorr);
\r
594 EXPECT_MAT_NEAR(dst, dst_gold, 1e-1);
\r
597 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Convolve, testing::Combine(
\r
600 testing::Values(KSize(3), KSize(7), KSize(11), KSize(17), KSize(19), KSize(23), KSize(45)),
\r
601 testing::Values(Ccorr(false), Ccorr(true))));
\r
603 ////////////////////////////////////////////////////////////////////////////////
\r
606 CV_ENUM(TemplateMethod, cv::TM_SQDIFF, cv::TM_SQDIFF_NORMED, cv::TM_CCORR, cv::TM_CCORR_NORMED, cv::TM_CCOEFF, cv::TM_CCOEFF_NORMED)
\r
607 #define ALL_TEMPLATE_METHODS testing::Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_SQDIFF_NORMED), TemplateMethod(cv::TM_CCORR), TemplateMethod(cv::TM_CCORR_NORMED), TemplateMethod(cv::TM_CCOEFF), TemplateMethod(cv::TM_CCOEFF_NORMED))
\r
609 IMPLEMENT_PARAM_CLASS(TemplateSize, cv::Size);
\r
611 PARAM_TEST_CASE(MatchTemplate8U, cv::gpu::DeviceInfo, cv::Size, TemplateSize, Channels, TemplateMethod)
\r
613 cv::gpu::DeviceInfo devInfo;
\r
615 cv::Size templ_size;
\r
619 virtual void SetUp()
\r
621 devInfo = GET_PARAM(0);
\r
622 size = GET_PARAM(1);
\r
623 templ_size = GET_PARAM(2);
\r
625 method = GET_PARAM(4);
\r
627 cv::gpu::setDevice(devInfo.deviceID());
\r
631 TEST_P(MatchTemplate8U, Accuracy)
\r
633 cv::Mat image = randomMat(size, CV_MAKETYPE(CV_8U, cn));
\r
634 cv::Mat templ = randomMat(templ_size, CV_MAKETYPE(CV_8U, cn));
\r
636 cv::gpu::GpuMat dst;
\r
637 cv::gpu::matchTemplate(loadMat(image), loadMat(templ), dst, method);
\r
640 cv::matchTemplate(image, templ, dst_gold, method);
\r
642 EXPECT_MAT_NEAR(dst_gold, dst, templ_size.area() * 1e-1);
\r
645 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate8U, testing::Combine(
\r
648 testing::Values(TemplateSize(cv::Size(5, 5)), TemplateSize(cv::Size(16, 16)), TemplateSize(cv::Size(30, 30))),
\r
649 testing::Values(Channels(1), Channels(3), Channels(4)),
\r
650 ALL_TEMPLATE_METHODS));
\r
652 ////////////////////////////////////////////////////////////////////////////////
\r
653 // MatchTemplate32F
\r
655 PARAM_TEST_CASE(MatchTemplate32F, cv::gpu::DeviceInfo, cv::Size, TemplateSize, Channels, TemplateMethod)
\r
657 cv::gpu::DeviceInfo devInfo;
\r
659 cv::Size templ_size;
\r
665 virtual void SetUp()
\r
667 devInfo = GET_PARAM(0);
\r
668 size = GET_PARAM(1);
\r
669 templ_size = GET_PARAM(2);
\r
671 method = GET_PARAM(4);
\r
673 cv::gpu::setDevice(devInfo.deviceID());
\r
677 TEST_P(MatchTemplate32F, Regression)
\r
679 cv::Mat image = randomMat(size, CV_MAKETYPE(CV_32F, cn));
\r
680 cv::Mat templ = randomMat(templ_size, CV_MAKETYPE(CV_32F, cn));
\r
682 cv::gpu::GpuMat dst;
\r
683 cv::gpu::matchTemplate(loadMat(image), loadMat(templ), dst, method);
\r
686 cv::matchTemplate(image, templ, dst_gold, method);
\r
688 EXPECT_MAT_NEAR(dst_gold, dst, templ_size.area() * 1e-1);
\r
691 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate32F, testing::Combine(
\r
694 testing::Values(TemplateSize(cv::Size(5, 5)), TemplateSize(cv::Size(16, 16)), TemplateSize(cv::Size(30, 30))),
\r
695 testing::Values(Channels(1), Channels(3), Channels(4)),
\r
696 testing::Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_CCORR))));
\r
698 ////////////////////////////////////////////////////////////////////////////////
\r
699 // MatchTemplateBlackSource
\r
701 PARAM_TEST_CASE(MatchTemplateBlackSource, cv::gpu::DeviceInfo, TemplateMethod)
\r
703 cv::gpu::DeviceInfo devInfo;
\r
706 virtual void SetUp()
\r
708 devInfo = GET_PARAM(0);
\r
709 method = GET_PARAM(1);
\r
711 cv::gpu::setDevice(devInfo.deviceID());
\r
715 TEST_P(MatchTemplateBlackSource, Accuracy)
\r
717 cv::Mat image = readImage("matchtemplate/black.png");
\r
718 ASSERT_FALSE(image.empty());
\r
720 cv::Mat pattern = readImage("matchtemplate/cat.png");
\r
721 ASSERT_FALSE(pattern.empty());
\r
723 cv::gpu::GpuMat d_dst;
\r
724 cv::gpu::matchTemplate(loadMat(image), loadMat(pattern), d_dst, method);
\r
726 cv::Mat dst(d_dst);
\r
730 cv::minMaxLoc(dst, NULL, &maxValue, NULL, &maxLoc);
\r
732 cv::Point maxLocGold = cv::Point(284, 12);
\r
734 ASSERT_EQ(maxLocGold, maxLoc);
\r
737 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplateBlackSource, testing::Combine(
\r
739 testing::Values(TemplateMethod(cv::TM_CCOEFF_NORMED), TemplateMethod(cv::TM_CCORR_NORMED))));
\r
741 ////////////////////////////////////////////////////////////////////////////////
\r
742 // MatchTemplate_CCOEF_NORMED
\r
744 PARAM_TEST_CASE(MatchTemplate_CCOEF_NORMED, cv::gpu::DeviceInfo, std::pair<std::string, std::string>)
\r
746 cv::gpu::DeviceInfo devInfo;
\r
747 std::string imageName;
\r
748 std::string patternName;
\r
750 virtual void SetUp()
\r
752 devInfo = GET_PARAM(0);
\r
753 imageName = GET_PARAM(1).first;
\r
754 patternName = GET_PARAM(1).second;
\r
756 cv::gpu::setDevice(devInfo.deviceID());
\r
760 TEST_P(MatchTemplate_CCOEF_NORMED, Accuracy)
\r
762 cv::Mat image = readImage(imageName);
\r
763 ASSERT_FALSE(image.empty());
\r
765 cv::Mat pattern = readImage(patternName);
\r
766 ASSERT_FALSE(pattern.empty());
\r
768 cv::gpu::GpuMat d_dst;
\r
769 cv::gpu::matchTemplate(loadMat(image), loadMat(pattern), d_dst, CV_TM_CCOEFF_NORMED);
\r
771 cv::Mat dst(d_dst);
\r
773 cv::Point minLoc, maxLoc;
\r
774 double minVal, maxVal;
\r
775 cv::minMaxLoc(dst, &minVal, &maxVal, &minLoc, &maxLoc);
\r
778 cv::matchTemplate(image, pattern, dstGold, CV_TM_CCOEFF_NORMED);
\r
780 double minValGold, maxValGold;
\r
781 cv::Point minLocGold, maxLocGold;
\r
782 cv::minMaxLoc(dstGold, &minValGold, &maxValGold, &minLocGold, &maxLocGold);
\r
784 ASSERT_EQ(minLocGold, minLoc);
\r
785 ASSERT_EQ(maxLocGold, maxLoc);
\r
786 ASSERT_LE(maxVal, 1.0);
\r
787 ASSERT_GE(minVal, -1.0);
\r
790 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate_CCOEF_NORMED, testing::Combine(
\r
792 testing::Values(std::make_pair(std::string("matchtemplate/source-0.png"), std::string("matchtemplate/target-0.png")))));
\r
794 ////////////////////////////////////////////////////////////////////////////////
\r
795 // MatchTemplate_CanFindBigTemplate
\r
797 struct MatchTemplate_CanFindBigTemplate : testing::TestWithParam<cv::gpu::DeviceInfo>
\r
799 cv::gpu::DeviceInfo devInfo;
\r
801 virtual void SetUp()
\r
803 devInfo = GetParam();
\r
805 cv::gpu::setDevice(devInfo.deviceID());
\r
809 TEST_P(MatchTemplate_CanFindBigTemplate, SQDIFF_NORMED)
\r
811 cv::Mat scene = readImage("matchtemplate/scene.jpg");
\r
812 ASSERT_FALSE(scene.empty());
\r
814 cv::Mat templ = readImage("matchtemplate/template.jpg");
\r
815 ASSERT_FALSE(templ.empty());
\r
817 cv::gpu::GpuMat d_result;
\r
818 cv::gpu::matchTemplate(loadMat(scene), loadMat(templ), d_result, CV_TM_SQDIFF_NORMED);
\r
820 cv::Mat result(d_result);
\r
824 cv::minMaxLoc(result, &minVal, 0, &minLoc, 0);
\r
826 ASSERT_GE(minVal, 0);
\r
827 ASSERT_LT(minVal, 1e-3);
\r
828 ASSERT_EQ(344, minLoc.x);
\r
829 ASSERT_EQ(0, minLoc.y);
\r
832 TEST_P(MatchTemplate_CanFindBigTemplate, SQDIFF)
\r
834 cv::Mat scene = readImage("matchtemplate/scene.jpg");
\r
835 ASSERT_FALSE(scene.empty());
\r
837 cv::Mat templ = readImage("matchtemplate/template.jpg");
\r
838 ASSERT_FALSE(templ.empty());
\r
840 cv::gpu::GpuMat d_result;
\r
841 cv::gpu::matchTemplate(loadMat(scene), loadMat(templ), d_result, CV_TM_SQDIFF);
\r
843 cv::Mat result(d_result);
\r
847 cv::minMaxLoc(result, &minVal, 0, &minLoc, 0);
\r
849 ASSERT_GE(minVal, 0);
\r
850 ASSERT_EQ(344, minLoc.x);
\r
851 ASSERT_EQ(0, minLoc.y);
\r
854 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate_CanFindBigTemplate, ALL_DEVICES);
\r
856 ////////////////////////////////////////////////////////////////////////////
\r
859 CV_FLAGS(DftFlags, 0, cv::DFT_INVERSE, cv::DFT_SCALE, cv::DFT_ROWS, cv::DFT_COMPLEX_OUTPUT, cv::DFT_REAL_OUTPUT)
\r
861 PARAM_TEST_CASE(MulSpectrums, cv::gpu::DeviceInfo, cv::Size, DftFlags)
\r
863 cv::gpu::DeviceInfo devInfo;
\r
869 virtual void SetUp()
\r
871 devInfo = GET_PARAM(0);
\r
872 size = GET_PARAM(1);
\r
873 flag = GET_PARAM(2);
\r
875 cv::gpu::setDevice(devInfo.deviceID());
\r
877 a = randomMat(size, CV_32FC2);
\r
878 b = randomMat(size, CV_32FC2);
\r
882 TEST_P(MulSpectrums, Simple)
\r
885 cv::gpu::mulSpectrums(loadMat(a), loadMat(b), c, flag, false);
\r
888 cv::mulSpectrums(a, b, c_gold, flag, false);
\r
890 EXPECT_MAT_NEAR(c_gold, c, 1e-2);
\r
893 TEST_P(MulSpectrums, Scaled)
\r
895 float scale = 1.f / size.area();
\r
898 cv::gpu::mulAndScaleSpectrums(loadMat(a), loadMat(b), c, flag, scale, false);
\r
901 cv::mulSpectrums(a, b, c_gold, flag, false);
\r
902 c_gold.convertTo(c_gold, c_gold.type(), scale);
\r
904 EXPECT_MAT_NEAR(c_gold, c, 1e-2);
\r
907 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MulSpectrums, testing::Combine(
\r
910 testing::Values(DftFlags(0), DftFlags(cv::DFT_ROWS))));
\r
912 ////////////////////////////////////////////////////////////////////////////
\r
915 struct Dft : testing::TestWithParam<cv::gpu::DeviceInfo>
\r
917 cv::gpu::DeviceInfo devInfo;
\r
919 virtual void SetUp()
\r
921 devInfo = GetParam();
\r
923 cv::gpu::setDevice(devInfo.deviceID());
\r
927 void testC2C(const std::string& hint, int cols, int rows, int flags, bool inplace)
\r
929 SCOPED_TRACE(hint);
\r
931 cv::Mat a = randomMat(cv::Size(cols, rows), CV_32FC2, 0.0, 10.0);
\r
934 cv::dft(a, b_gold, flags);
\r
936 cv::gpu::GpuMat d_b;
\r
937 cv::gpu::GpuMat d_b_data;
\r
940 d_b_data.create(1, a.size().area(), CV_32FC2);
\r
941 d_b = cv::gpu::GpuMat(a.rows, a.cols, CV_32FC2, d_b_data.ptr(), a.cols * d_b_data.elemSize());
\r
943 cv::gpu::dft(loadMat(a), d_b, cv::Size(cols, rows), flags);
\r
945 EXPECT_TRUE(!inplace || d_b.ptr() == d_b_data.ptr());
\r
946 ASSERT_EQ(CV_32F, d_b.depth());
\r
947 ASSERT_EQ(2, d_b.channels());
\r
948 EXPECT_MAT_NEAR(b_gold, cv::Mat(d_b), rows * cols * 1e-4);
\r
953 int cols = randomInt(2, 100);
\r
954 int rows = randomInt(2, 100);
\r
956 for (int i = 0; i < 2; ++i)
\r
958 bool inplace = i != 0;
\r
960 testC2C("no flags", cols, rows, 0, inplace);
\r
961 testC2C("no flags 0 1", cols, rows + 1, 0, inplace);
\r
962 testC2C("no flags 1 0", cols, rows + 1, 0, inplace);
\r
963 testC2C("no flags 1 1", cols + 1, rows, 0, inplace);
\r
964 testC2C("DFT_INVERSE", cols, rows, cv::DFT_INVERSE, inplace);
\r
965 testC2C("DFT_ROWS", cols, rows, cv::DFT_ROWS, inplace);
\r
966 testC2C("single col", 1, rows, 0, inplace);
\r
967 testC2C("single row", cols, 1, 0, inplace);
\r
968 testC2C("single col inversed", 1, rows, cv::DFT_INVERSE, inplace);
\r
969 testC2C("single row inversed", cols, 1, cv::DFT_INVERSE, inplace);
\r
970 testC2C("single row DFT_ROWS", cols, 1, cv::DFT_ROWS, inplace);
\r
971 testC2C("size 1 2", 1, 2, 0, inplace);
\r
972 testC2C("size 2 1", 2, 1, 0, inplace);
\r
976 void testR2CThenC2R(const std::string& hint, int cols, int rows, bool inplace)
\r
978 SCOPED_TRACE(hint);
\r
980 cv::Mat a = randomMat(cv::Size(cols, rows), CV_32FC1, 0.0, 10.0);
\r
982 cv::gpu::GpuMat d_b, d_c;
\r
983 cv::gpu::GpuMat d_b_data, d_c_data;
\r
988 d_b_data.create(1, (a.rows / 2 + 1) * a.cols, CV_32FC2);
\r
989 d_b = cv::gpu::GpuMat(a.rows / 2 + 1, a.cols, CV_32FC2, d_b_data.ptr(), a.cols * d_b_data.elemSize());
\r
993 d_b_data.create(1, a.rows * (a.cols / 2 + 1), CV_32FC2);
\r
994 d_b = cv::gpu::GpuMat(a.rows, a.cols / 2 + 1, CV_32FC2, d_b_data.ptr(), (a.cols / 2 + 1) * d_b_data.elemSize());
\r
996 d_c_data.create(1, a.size().area(), CV_32F);
\r
997 d_c = cv::gpu::GpuMat(a.rows, a.cols, CV_32F, d_c_data.ptr(), a.cols * d_c_data.elemSize());
\r
1000 cv::gpu::dft(loadMat(a), d_b, cv::Size(cols, rows), 0);
\r
1001 cv::gpu::dft(d_b, d_c, cv::Size(cols, rows), cv::DFT_REAL_OUTPUT | cv::DFT_SCALE);
\r
1003 EXPECT_TRUE(!inplace || d_b.ptr() == d_b_data.ptr());
\r
1004 EXPECT_TRUE(!inplace || d_c.ptr() == d_c_data.ptr());
\r
1005 ASSERT_EQ(CV_32F, d_c.depth());
\r
1006 ASSERT_EQ(1, d_c.channels());
\r
1009 EXPECT_MAT_NEAR(a, c, rows * cols * 1e-5);
\r
1012 TEST_P(Dft, R2CThenC2R)
\r
1014 int cols = randomInt(2, 100);
\r
1015 int rows = randomInt(2, 100);
\r
1017 testR2CThenC2R("sanity", cols, rows, false);
\r
1018 testR2CThenC2R("sanity 0 1", cols, rows + 1, false);
\r
1019 testR2CThenC2R("sanity 1 0", cols + 1, rows, false);
\r
1020 testR2CThenC2R("sanity 1 1", cols + 1, rows + 1, false);
\r
1021 testR2CThenC2R("single col", 1, rows, false);
\r
1022 testR2CThenC2R("single col 1", 1, rows + 1, false);
\r
1023 testR2CThenC2R("single row", cols, 1, false);
\r
1024 testR2CThenC2R("single row 1", cols + 1, 1, false);
\r
1026 testR2CThenC2R("sanity", cols, rows, true);
\r
1027 testR2CThenC2R("sanity 0 1", cols, rows + 1, true);
\r
1028 testR2CThenC2R("sanity 1 0", cols + 1, rows, true);
\r
1029 testR2CThenC2R("sanity 1 1", cols + 1, rows + 1, true);
\r
1030 testR2CThenC2R("single row", cols, 1, true);
\r
1031 testR2CThenC2R("single row 1", cols + 1, 1, true);
\r
1034 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Dft, ALL_DEVICES);
\r
1036 ///////////////////////////////////////////////////////////////////////////////////////////////////////
\r
1039 IMPLEMENT_PARAM_CLASS(BlockSize, int);
\r
1040 IMPLEMENT_PARAM_CLASS(ApertureSize, int);
\r
1042 PARAM_TEST_CASE(CornerHarris, cv::gpu::DeviceInfo, MatType, BorderType, BlockSize, ApertureSize)
\r
1044 cv::gpu::DeviceInfo devInfo;
\r
1050 virtual void SetUp()
\r
1052 devInfo = GET_PARAM(0);
\r
1053 type = GET_PARAM(1);
\r
1054 borderType = GET_PARAM(2);
\r
1055 blockSize = GET_PARAM(3);
\r
1056 apertureSize = GET_PARAM(4);
\r
1058 cv::gpu::setDevice(devInfo.deviceID());
\r
1062 TEST_P(CornerHarris, Accuracy)
\r
1064 cv::Mat src = readImageType("stereobm/aloe-L.png", type);
\r
1065 ASSERT_FALSE(src.empty());
\r
1067 double k = randomDouble(0.1, 0.9);
\r
1069 cv::gpu::GpuMat dst;
\r
1070 cv::gpu::cornerHarris(loadMat(src), dst, blockSize, apertureSize, k, borderType);
\r
1073 cv::cornerHarris(src, dst_gold, blockSize, apertureSize, k, borderType);
\r
1075 EXPECT_MAT_NEAR(dst_gold, dst, 0.02);
\r
1078 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CornerHarris, testing::Combine(
\r
1080 testing::Values(MatType(CV_8UC1), MatType(CV_32FC1)),
\r
1081 testing::Values(BorderType(cv::BORDER_REFLECT101), BorderType(cv::BORDER_REPLICATE), BorderType(cv::BORDER_REFLECT)),
\r
1082 testing::Values(BlockSize(3), BlockSize(5), BlockSize(7)),
\r
1083 testing::Values(ApertureSize(0), ApertureSize(3), ApertureSize(5), ApertureSize(7))));
\r
1085 ///////////////////////////////////////////////////////////////////////////////////////////////////////
\r
1088 PARAM_TEST_CASE(CornerMinEigen, cv::gpu::DeviceInfo, MatType, BorderType, BlockSize, ApertureSize)
\r
1090 cv::gpu::DeviceInfo devInfo;
\r
1096 virtual void SetUp()
\r
1098 devInfo = GET_PARAM(0);
\r
1099 type = GET_PARAM(1);
\r
1100 borderType = GET_PARAM(2);
\r
1101 blockSize = GET_PARAM(3);
\r
1102 apertureSize = GET_PARAM(4);
\r
1104 cv::gpu::setDevice(devInfo.deviceID());
\r
1108 TEST_P(CornerMinEigen, Accuracy)
\r
1110 cv::Mat src = readImageType("stereobm/aloe-L.png", type);
\r
1111 ASSERT_FALSE(src.empty());
\r
1113 cv::gpu::GpuMat dst;
\r
1114 cv::gpu::cornerMinEigenVal(loadMat(src), dst, blockSize, apertureSize, borderType);
\r
1117 cv::cornerMinEigenVal(src, dst_gold, blockSize, apertureSize, borderType);
\r
1119 EXPECT_MAT_NEAR(dst_gold, dst, 0.02);
\r
1122 INSTANTIATE_TEST_CASE_P(GPU_ImgProc, CornerMinEigen, testing::Combine(
\r
1124 testing::Values(MatType(CV_8UC1), MatType(CV_32FC1)),
\r
1125 testing::Values(BorderType(cv::BORDER_REFLECT101), BorderType(cv::BORDER_REPLICATE), BorderType(cv::BORDER_REFLECT)),
\r
1126 testing::Values(BlockSize(3), BlockSize(5), BlockSize(7)),
\r
1127 testing::Values(ApertureSize(0), ApertureSize(3), ApertureSize(5), ApertureSize(7))));
\r
1131 #endif // HAVE_CUDA
\r