Research map / Detection and segmentation / CNN features
ECO-TR: Efficient Correspondences Finding via Coarse-to-Fine Refinement
Dongli Tan, Jiang-Jiang Liu, Xingyu Chen, Chao Chen, Ruixin Zhang, Yunhang Shen, Shouhong Ding, Rongrong Ji
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 |
|---|---|---|---|
| add middle stage to coarse-only model (ablation) | ECO-TR · ETH3D rate=15 | AEPE: 9.19 -> 3.83 | Improved |
| add fine stage to coarse+middle model (ablation) | ECO-TR · ETH3D rate=15 | AEPE: 3.83 -> 2.72 | Improved |
| add deep supervision of middle and fine branches during training (ablation) | ECO-TR · ETH3D rate=15 | AEPE: 9.53 -> 9.19 | Improved |
| use adaptive query-clustering instead of GRID clustering (ablation) | ECO-TR · ETH3D rate=15 | AEPE: 3.10 -> 2.72 | Improved |
| use adaptive query-clustering instead of GRID clustering (ablation) | ECO-TR · ETH3D rate=3 | AEPE: 1.64 -> 1.53 | Improved |
| replace full attention with linear transformer in middle and fine stages (ablation) | ECO-TR · ETH3D rate=3 | AEPE: 1.53 -> 1.55 | Hurt |
| use uncertainty filter instead of cycle consistency filter (ablation) | ECO-TR · ETH3D rate=15 | AEPE: 2.45 -> 2.08 | Improved |
| combine uncertainty and cycle consistency filtering (ablation) | ECO-TR · ETH3D rate=15 | AEPE: 2.08 -> 2.06 | Improved |
| end-to-end feature-level cropping vs recursive zoom-in of COTR (10k queries) | ECO-TR · 10k queries | inference time (sec): 433.53 -> 2.35 | Improved |
| ECO-TR vs COTR on HPatches (sampled matches) | ECO-TR · HPatches | PCK-5px: 90.85 -> 90.71 | Hurt |
| ECO-TR vs COTR on KITTI-2015 | ECO-TR · KITTI-2015 | AEPE: 2.06 -> 1.40 | Improved |
| ECO-TR vs COTR on MegaDepth with 300 matches | ECO-TR · MegaDepth St. Paul's (N=300) | mAA@5deg: 0.434 -> 0.430 | No clear effect |
Closest papers by meaning
- COTR: Correspondence Transformer for Matching Across Images (ICCV 2021 · similarity 0.87)
- ETO:Efficient Transformer-based Local Feature Matching by Organizing Multiple Homography Hypotheses (NeurIPS 2024 · similarity 0.86)
- Diffusion Model for Dense Matching (ICLR 2024 · similarity 0.85)
- Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed (CVPR 2024 · similarity 0.85)
- MATCHA: Towards Matching Anything (CVPR 2025 · similarity 0.84)
- HomoMatcher: Achieving Dense Feature Matching with Semi-Dense Efficiency by Homography Estimation (AAAI 2025 · similarity 0.84)
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.
