Research: LoTex
LoTex is a controlled failure-oriented evaluation of 17 feature matching methods under geometric and photometric stress on low-texture planar scenes.
We published LoTex, a failure-oriented empirical analysis of feature matching pipelines on low-texture planar scenes.
Unlike leaderboard-style benchmarks, LoTex focuses on how and why matching methods degrade under controlled transformations such as rotation, blur, tilt, and combined camera motion effects.
The evaluation includes:
- 250 real low-texture planar images
- 12 transformation types x 3 difficulty levels
- 9,000 image pairs with known ground-truth homographies
- 17 classical and learning-based matching methods
- Homography-based robustness and failure analysis
The results reveal systematic robustness gaps under rotations, blur, and combined degradations, as well as qualitative differences between gradual geometric degradation and abrupt pipeline failures.
LoTex serves as a diagnostic framework for understanding matching behavior beyond aggregate success metrics.