// This file is part of OpenCV project. // It is subject to the license terms in the LICENSE file found in the top-level directory // of this distribution and at http://opencv.org/license.html. #include "test_precomp.hpp" #include "npy_blob.hpp" #ifdef HAVE_OPENCV_DNN #include "opencv2/dnn.hpp" #include "opencv2/core/utils/logger.hpp" #endif namespace opencv_test { namespace { #ifdef HAVE_OPENCV_DNN TEST(Features2d_LightGlue_DISK, Regression) { applyTestTag(CV_TEST_TAG_MEMORY_2GB); const std::string diskPath = cvtest::findDataFile("dnn/disk.onnx", false); const std::string lgPath = cvtest::findDataFile("dnn/onnx/models/disk_lightglue.onnx", false); Ptr disk = DISK::create(diskPath); Ptr lg = LightGlueMatcher::create(lgPath, 0.0f, 0, 0, LG_DISK); ASSERT_FALSE(disk.empty()); ASSERT_FALSE(lg.empty()); // Limit keypoints to prevent OOM (LightGlue Transformer self-attention is O(N²)) disk->setMaxKeypoints(1024); Mat img1 = imread(cvtest::findDataFile("shared/box.png")); Mat img2 = imread(cvtest::findDataFile("shared/box_in_scene.png")); ASSERT_FALSE(img1.empty()); ASSERT_FALSE(img2.empty()); // Detect features on both images std::vector kpts1, kpts2; Mat descs1, descs2; disk->detectAndCompute(img1, noArray(), kpts1, descs1); disk->detectAndCompute(img2, noArray(), kpts2, descs2); ASSERT_GT(static_cast(kpts1.size()), 0); ASSERT_GT(static_cast(kpts2.size()), 0); ASSERT_EQ(kpts1.size(), 1024u); // maxKeypoints applied ASSERT_EQ(kpts2.size(), 1024u); // Build keypoint matrices (pixel coordinates) Mat kpts1Mat(static_cast(kpts1.size()), 2, CV_32F); Mat kpts2Mat(static_cast(kpts2.size()), 2, CV_32F); for (size_t i = 0; i < kpts1.size(); i++) { kpts1Mat.at(static_cast(i), 0) = kpts1[i].pt.x; kpts1Mat.at(static_cast(i), 1) = kpts1[i].pt.y; } for (size_t i = 0; i < kpts2.size(); i++) { kpts2Mat.at(static_cast(i), 0) = kpts2[i].pt.x; kpts2Mat.at(static_cast(i), 1) = kpts2[i].pt.y; } lg->setPairInfo(kpts1Mat, kpts2Mat, img1.size(), img2.size()); std::vector matches; lg->match(descs1, descs2, matches); ASSERT_GT(static_cast(matches.size()), 0); for (const auto& m : matches) { EXPECT_GE(m.queryIdx, 0); EXPECT_LT(m.queryIdx, static_cast(kpts1.size())); EXPECT_GE(m.trainIdx, 0); EXPECT_LT(m.trainIdx, static_cast(kpts2.size())); } // Load ORT reference outputs Mat refMatches = blobFromNPY(cvtest::findDataFile("features/disk/disk_lightglue_matches.npy")); // Match count tolerance: ±2 to allow for ONNX Runtime version / floating-point differences ASSERT_LE(std::abs(static_cast(matches.size()) - refMatches.rows), 2) << "Match count mismatch: got " << matches.size() << ", expected " << refMatches.rows << " (±2 tolerance)"; // Compare match sets instead of ordered sequences. DNN engine differences // (e.g. new graph engine vs ORT) may produce different match ordering while // the actual matched pairs are equivalent. Allow up to 3 mismatches. int refNotFound = 0; for (int i = 0; i < refMatches.rows; i++) { int refQIdx = static_cast(refMatches.at(i, 0)); int refTIdx = static_cast(refMatches.at(i, 1)); bool found = false; for (const auto& m : matches) { if (m.queryIdx == refQIdx && m.trainIdx == refTIdx) { found = true; break; } } if (!found) refNotFound++; } int actNotFound = 0; for (const auto& m : matches) { bool found = false; for (int i = 0; i < refMatches.rows; i++) { int refQIdx = static_cast(refMatches.at(i, 0)); int refTIdx = static_cast(refMatches.at(i, 1)); if (m.queryIdx == refQIdx && m.trainIdx == refTIdx) { found = true; break; } } if (!found) actNotFound++; } EXPECT_LE(refNotFound, 3) << refNotFound << "/" << refMatches.rows << " reference matches not found in actual output"; EXPECT_LE(actNotFound, 3) << actNotFound << "/" << matches.size() << " actual matches not found in reference"; if (refNotFound > 0 || actNotFound > 0) CV_LOG_INFO(NULL, "DISK LightGlue: " << refNotFound << " ref missing, " << actNotFound << " actual extra (within tolerance)"); } #else // !HAVE_OPENCV_DNN TEST(Features2d_LightGlueMatcher_DISK, not_available) { EXPECT_THROW(LightGlueMatcher::create("dummy.onnx", 0.0f, 0, 0, LG_DISK), cv::Exception); } #endif // HAVE_OPENCV_DNN }} // namespace opencv_test