Research map / Detection and segmentation / CNN features
S-TREK: Sequential Translation and Rotation Equivariant Keypoints for Local Feature Extraction
Emanuele Santellani, Christian Sormann, Mattia Rossi, Andreas Kuhn, Friedrich Fraundorfer
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 |
|---|---|---|---|
| replace peaky+similarity heatmap losses with sequential-sampling RL-style training (ablation) | ReCNN detector · Synthetic lines validation | max repeatability@1px: 0.58 -> 0.78 | Improved |
| vary number of sequential samples (50, 100, 200) (ablation) | S-TREK detector · Synthetic lines validation | number of detected keypoints: 39, 53, 56 keypoints | No clear effect |
| vary peaky window size (32, 64, 96) (ablation) | peaky+similarity baseline · Synthetic lines validation | number of detected keypoints: 27 to 60 | Hurt |
| use rotation-equivariant convolutions instead of standard CNN with same channels (ablation) | S-TREK detector · Synthetic lines validation | max repeatability@1px: 0.61 -> 0.78 | Improved |
| use ReCNN instead of a much larger standard CNN (377k params) (ablation) | S-TREK detector · Synthetic lines validation | max repeatability@1px: 0.68 -> 0.78 | Improved |
| use a smaller shallow ReCNN (cyclic 8) detector (ablation) | S-TREK detector · HPatches rotation repeatability | mean repeatability | Hurt |
| increase cyclicity from 8 to 16 in the smaller ReCNN (ablation) | smaller ReCNN · HPatches rotation repeatability | mean repeatability | No clear effect |
| rotation-equivariant detector with detect-then-describe | S-TREK · HPatches +-45 deg | Homography accuracy AUC@3px: 0.182 -> 0.248 | Improved |
| S-TREK vs DISK on unrotated Image Matching Benchmark | S-TREK · Image Matching Benchmark standard | mAA@10: 0.512 vs 0.458 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- R2D2 — R2D2: Reliable and Repeatable Detector and Descriptor (NeurIPS 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- RIPE: Reinforcement Learning on Unlabeled Image Pairs for Robust Keypoint Extraction (ICCV 2025 · similarity 0.86)
- Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Task (CVPR 2020 · similarity 0.85)
- ESCAPE: Encoding Super-keypoints for Category-Agnostic Pose Estimation (CVPR 2024 · similarity 0.83)
- Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features (NeurIPS 2024 · similarity 0.83)
- Detect Any Keypoints: An Efficient Light-Weight Few-Shot Keypoint Detector (AAAI 2024 · similarity 0.83)
- SDGMNet: Statistic-Based Dynamic Gradient Modulation for Local Descriptor Learning (AAAI 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.
