Ground Truth Matches (GTM) to test computer vision algorithms which use image features

GTM for the datasets KITTI 2015 and Oxford (Mikolajczyk et al.) can be downloaded here.

For the corresponding image data please visit

A detailed description of the GTM and a few test results can be found in our paper. Further results on the performance of keypoint-descriptor pairings and matching algorithms can be downloaded here: Supplementary material (4360 downloads )

Additional information and an interface for reading and viewing GTM can be found at https://github.com/josefmaierfl/GTM

Downloads:

Complete GTMs for KITTI 2015 and Oxford datasets including all available keypoint types and inlier ratios:

GTMs for KITTI 2015 disparity dataset and various inlier ratios ranging from 1% to 100%:

GTMs for KITTI 2015 disparity dataset and no specif inlier ratios providing the highest number of matches and true negatives:

GTM for KITTI 2015 disparity with all available keypoints - no specific inlier ratio (2391 downloads )

GTMs for KITTI 2015 flow dataset and various inlier ratios ranging from 1% to 100%:

GTMs for KITTI 2015 disparity dataset and no specif inlier ratios providing the highest number of matches and true negatives:

GTM for KITTI 2015 flow with all available keypoints - no specific inlier ratio (2300 downloads )

GTMs for Oxford (Mikolajczyk et al.) dataset and various inlier ratios ranging from 1% to 100%:

GTMs for Oxford (Mikolajczyk et al.) dataset and no specif inlier ratios providing the highest number of matches and true negatives: