Research map / Image generation / Image restoration
Affine steerers for structured keypoint description
Georg Bökman, Johan Edstedt, Michael Felsberg, Fredrik Kahl
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain with affine steerers on homographies, then fine-tune on upright images with learned max-similarity prototypes | DeDoDe-B · IMC2022 hidden test | mAA@10: 72.9 -> 77.3 | Improved |
| same pipeline applied to DeDoDe-B on MegaDepth-1500 | DeDoDe-B · MegaDepth-1500 relative pose | AUC@5deg: 49.4 -> 52.7 | Improved |
| same pipeline applied to DeDoDe-G | DeDoDe-G · HPatches homography | AUC@3px: 67.1 -> 68.2 | Improved |
| use AffSteer-G instead of DeDoDe-G on WxBS | DeDoDe-G · WxBS fundamental matrix | mAA@10px: 57.7 -> 57.3 | No clear effect |
| use AffEqui-B (equivariant) instead of upright-specialized models on MegaDepth-1500 | DeDoDe-B · MegaDepth-1500 relative pose | AUC@5deg: 49.4 -> 46.1 | Hurt |
| add a steerer when training with heavy affine augmentation (ablation) | DeDoDe-B · MegaDepth-1500 relative pose | AUC@5deg: 40.8 -> 46.1 | Improved |
| add homography pretraining to AffSteer (ablation) | AffSteer-B · MegaDepth-1500 relative pose | AUC@5deg: 47.2 -> 52.7 | Improved |
| add homography pretraining to AffEqui (ablation) | AffEqui-B · MegaDepth-1500 relative pose | AUC@5deg: 46.1 -> 47.5 | Improved |
| train XFeat descriptor with the affine steerer pretraining and max-similarity fine-tuning pipeline | XFeat · MegaDepth-1500 relative pose | AUC@5deg: 43.2 -> 48.6 | Improved |
| increase max similarity rotations from 8 to 12 (ablation) | AffEqui-B · AIMS (All) | average precision: 59 -> 60 | Improved |
| fine-tune descriptors on upright images with max similarity | AffSteer-B · affine steering oracle | steering quality | Hurt |
Closest papers by meaning
- Steerers: A Framework for Rotation Equivariant Keypoint Descriptors (CVPR 2024 · similarity 0.89)
- Learning 3D Keypoint Descriptors for Non-Rigid Shape Matching (ECCV 2018 · similarity 0.86)
- RIPE: Reinforcement Learning on Unlabeled Image Pairs for Robust Keypoint Extraction (ICCV 2025 · similarity 0.85)
- Learning Affine Correspondences by Integrating Geometric Constraints (CVPR 2025 · similarity 0.84)
- KeyPose: Category-Level 6D Object Pose Estimation with Self-Adaptive Keypoints (AAAI 2025 · similarity 0.83)
- Meta-Point Learning and Refining for Category-Agnostic Pose Estimation (CVPR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
