A case for using rotation invariant features in state of the art feature matchers
About
The aim of this paper is to demonstrate that a state of the art feature matcher (LoFTR) can be made more robust to rotations by simply replacing the backbone CNN with a steerable CNN which is equivariant to translations and image rotations. It is experimentally shown that this boost is obtained without reducing performance on ordinary illumination and viewpoint matching sequences.
Georg B\"okman, Fredrik Kahl• 2022
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Feature Matching | HPatches | AUC @3°67.11 | 36 | |
| Feature Matching | MegaDepth | AUC @5°47.88 | 18 | |
| Feature Matching | ScanNet | AUC @5°17.77 | 18 | |
| Feature Matching | MegaDepth-C4 | AUC @5°33.37 | 9 | |
| Feature Matching | ScanNet C4 | AUC @5°13.24 | 9 | |
| Feature Matching | SatAst | AUC @10°0.22 | 9 | |
| Image Matching | AIMS (North Up) | AP58 | 4 | |
| Image Matching | AIMS (All Others) | AP51 | 4 | |
| Image Matching | AIMS (All) | AP52 | 4 |
Showing 9 of 9 rows