UM-Homo-Dataset provides some image pairs and feature point pairs on them for evaluating the accuracy of Homography matrix estimation.
The image data of UM-Homo-Dataset is collected by robots in real scenes, which have various lighting changes and dynamic objects.
UM-Homo-Dataset provides groundtruth feature point pairs that are preprocessed with SIFT and then manually selected, and each pair of images has ten pairs of relatively evenly distributed feature points.
Select about 500 image pairs from the raw data of UM-CAD-Dataset, mark the feature point pairs, and use it to evaluate the algorithm.
- SIFT+RANSAC
- SIFT+MAGSAC
- ORB+RANSAC
- ORB+MAGSAC
- BEBLID+RANSAC
- BEBLID+MAGSAC
- ours
- https://github.com/danini/magsac (Opencv?)