ALoMa: Affine Correspondences Beyond Patches
Organizations: Chalmers University of Technology · University of Amsterdam · Lund University · California Polytechnic State University
Abstract
Affine correspondences (ACs) enable robust and efficient solvers for multiple-view geometry by constraining the local geometry of point correspondences (PCs). Despite their utility in minimal solvers, AC estimation remains understudied. We revisit AC estimation beyond local image patches, extending it to the image and image-pair levels. Rather than estimating the components of the affine transformation separately, we learn the full 4-DoF matrix jointly. Building on this formulation, we introduce ALoMa, an AC estimator using LoMa keypoint descriptions, demonstrating consistent improvements across a wide range of downstream tasks. Code at https://github.com/davnords/loma.
Figures & tables
| Method | HPatches | Mega1500 | ||
|---|---|---|---|---|
| Frobenius | Rot. | Frobenius | Rot. | |
| Differentiated dense warp, at LoMa keypoints | ||||
| RoMa v2 | 2.82 | 0.46 ∘ | 5.04 | 0.75 ∘ |
| KN+HN+AN | 26.9 | 6.35 ∘ | 32.7 | 7.17 ∘ |
| SIFT+HN+S3E | 54.4 | 9.52 ∘ | 49.3 | 5.50 ∘ |
| Sun et al. | 48.2 | 3.82 ∘ | 40.1 | 2.09 ∘ |
| Mega1500 | ScanNet1500 | |||||
| Method | ||||||
| KN+HN+AN | 46.4 | 58.7 | 67.5 | 6.8 | 13.2 | 20.4 |
| SIFT+HN+S3E | 46.1 | 58.3 | 67.2 | 5.7 | 11.4 | 18.4 |
| Sun et al. † | 61.5 | 73.1 | 81.1 | 16.4 | 30.7 | 44.7 |
| Sun et al. ‡ | 65.5 | 77.9 | 86.3 | — | — | — |
| LoMa+S3E+AN | 63.4 | 77.2 | 86.7 | 28.1 | 50.0 | 67.9 |
| Solver | Method | @ 5 ° | @ 10 ° | @ 20 ° |
|---|---|---|---|---|
| 20PC | KN+HN | 3.0 | 5.1 | 7.8 |
| SIFT+HN | 11.1 | 14.6 | 17.2 | |
| DKM | 41.2 | 47.5 | 52.0 | |
| LoMa-B | 22.6 | 30.7 | 36.7 | |
| 7AC | KN+HN+AN | 36.9 | 46.2 | 53.3 |
| SIFT+HN+S3E | 37.6 | 44.4 | 49.6 |
| [email protected] m, | ||||
| Solver | Method | Iters. | No cap | 100 it. |
| 26PC | KN+HN | 930 | 75.7 | 61.2 |
| SIFT+HN | 911 | 76.2 | 61.2 | |
| DKM | 303 | 73.3 | 72.1 | |
| LoMa-B | 1009 | 78.7 | 64.7 | |
| 9AC | KN+HN+AN | 2288 | 56.3 | 19.3 |
| Solver | Method | 25 cm, | 50 cm, |
|---|---|---|---|
| P4Pf | Points only | 41.4 0.07 | 63.3 0.05 |
| UP1SIFT | S3Esti | 44.8 0.17 | 59.6 0.16 |
| LoMa-B | 49.9 0.15 | 65.5 0.14 | |
| UP1PfAC | S3Esti+AffNet | 39.5 0.15 | 61.5 0.13 |
| LoMa-B | 40.6 0.14 | 62.1 0.12 | |
| P1AC | S3Esti | 67.0 0.12 | 84.7 0.09 |
| Method | Map-free [ 7 ] | Mega1500 | ||||
|---|---|---|---|---|---|---|
| 5 ∘ | 10 ∘ | 20 ∘ | 5 ∘ | 10 ∘ | 20 ∘ | |
| Differentiated dense warp, at LoMa keypoints | ||||||
| RoMa v2 | 5.9 | 14.1 | 27.8 | 4.3 | 12.7 | 29.2 |
| KN+HN+AN | 0.3 | 1.3 | 4.8 | 0.3 | 1.3 | 4.4 |
| SIFT+HN+S3E | 0.2 | 0.7 | 2.8 | 0.1 | 0.5 | 1.9 |
| Sun et al. | 0.1 | 0.5 | 1.9 | 0.2 | 0.9 | 3.3 |
| Initialization | Matcher | Affine head | Frob. | AUC | |
| A | Scratch | Linear | 4.6 | 83.2 | |
| B | LoMa | Linear | 4.8 | 83.2 | |
| C | LoMa | ❆ | Linear | 12.2 | 83.6 |
| D | Scratch | Transformer | 3.8 | 83.2 | |
| E | LoMa | ❆ | Transformer | 4.0 | 83.6 |
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
| Datasets | Type / GT Source | Weight |
|---|---|---|
| AerialMD [ 92 ] | Aerial / MVS | |
| ASE [ 9 ] | Indoor / Graphics | |
| BlendedMVS [ 100 ] | Aerial / Mesh | |
| FlyingThings3D [ 61 ] | Synthetic / Graphics | |
| Hypersim [ 73 ] | Indoor / Graphics | |
| Map-free [ 7 ] | Object-centric / MVS |
| Solver | Method | 25 cm, | 50 cm, |
|---|---|---|---|
| P4Pf | Points only | 41.4 0.04 | 63.7 0.02 |
| UP1SIFT | S3Esti | 45.0 0.15 | 59.7 0.15 |
| LoMa-B | 49.8 0.15 | 65.5 0.16 | |
| UP1PfAC | S3Esti+AffNet | 44.7 0.07 | 66.0 0.06 |
| LoMa-B | 44.7 0.06 | 66.1 0.07 | |
| P1AC | S3Esti | 67.1 0.13 | 84.7 0.10 |
| Method | Throughput (pairs / second) |
|---|---|
| Sun et al. | 3.9 |
| RoMa v2 | 4.3 |
| LoMa-B | 51.1 |
| LoMa-L | 40.5 |
| LoMa-G | 35.7 |
| Method | HPatches | Mega1500 | ||||
|---|---|---|---|---|---|---|
| Point | Point | |||||
| RoMa v2 | 96.4 | 94.5 | 89.3 | 72.0 | 63.1 | 47.8 |
| KN+HN+AN | 39.1 | 17.7 | 2.8 | 8.7 | 3.0 | 0.5 |
| SIFT+HN+S3E | 75.0 | 13.2 | 2.0 | 12.7 | 3.0 | 0.7 |
| Sun et al. | 66.5 | 11.8 | 2.0 | 57.1 | 17.4 | 4.9 |
| LoMa+S3E+AN | 51.2 | 9.8 | 1.7 | 29.1 | 9.1 | 2.8 |
| Method | HPatches | Mega1500 | ||
|---|---|---|---|---|
| Frobenius | Rot. | Frobenius | Rot. | |
| KN+HN+AN | 25.8 | 6.03 ∘ | 34.3 | 7.66 ∘ |
| SIFT+HN+S3E | 51.2 | 8.83 ∘ | 48.8 | 5.24 ∘ |
| Sun et al. | 47.3 | 3.39 ∘ | 40.1 | 2.00 ∘ |
| LoMa+S3E+AN | 50.0 | 4.46 ∘ | 42.9 | 2.40 ∘ |
| LoMa-B | 1.5 | 0.25 ∘ | 4.1 | 0.61 ∘ |
| (px) | 0.25 | 0.5 | 1 | 2 | 4 | 8 |
|---|---|---|---|---|---|---|
| Median | 0.007 | 0.014 | 0.027 | 0.054 | 0.108 | 0.217 |
| Mean | 0.010 | 0.021 | 0.042 | 0.083 | 0.167 | 0.334 |
| 95th pct. | 0.031 | 0.062 | 0.124 | 0.248 | 0.496 | 0.993 |
| HPatches | Mega1500 | |||||||
| Aff. Frob. | Aff. Frob. | AUC@ | ||||||
| Matcher | ❆ / | RA | ||||||
| LoMa-B | ❆ | ✗ | 1.62 | 3.30 | 3.97 | 7.01 | 83.6 | 81.2 |
| LoMa-R | ❆ | ✓ | 8.00 | 9.01 | 8.74 | 8.96 | 82.8 | 83.2 |
| LoMa-R | ✓ | 1.69 | 1.83 | 4.06 | 4.11 | 83.4 | 82.8 | |
| LoMa-B | ✓ | 1.58 | 1.70 | 4.27 | 4.58 | 82.8 | 82.6 | |
Multi-View Geometry
Multi-view geometry studies how images from different viewpoints reveal camera motion, scene structure, and consistent 3D measurements.
See all 111 papers