Research map / 3D vision / Human pose estimation
Improving Transformer-based Image Matching by Cascaded Capturing Spatially Informative Keypoints
Chenjie Cao, Yanwei Fu
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 trivial dense refinement with cascade matching (ablation) | transformer matcher baseline · MegaDepth | AUC@5deg: 47.09 -> 57.99 (with NMS 5) | Improved |
| add NMS detection to CasMTR-4c (ablation) | CasMTR-4c · MegaDepth | AUC@5deg: 56.34 -> 57.99 | Improved |
| add NMS to baseline without cascade matching (ablation) | baseline matcher · MegaDepth | AUC@5deg: 55.77 -> 55.41 | No clear effect |
| add 1/2 resolution cascade to 4c with NMS 5 (ablation) | CasMTR · MegaDepth | AUC@5deg: 57.99 -> 59.08 | Improved |
| use CasMTR-2c instead of CasMTR-4c (ablation) | CasMTR · ScanNet | AUC | No clear effect |
| raise confidence threshold from 0.2 to 0.5 instead of NMS (ablation) | CasMTR-4c · MegaDepth | AUC@5deg: 56.34 -> 53.81 | Hurt |
| replace NMS with trainable detector (ablation) | CasMTR-4c · MegaDepth | AUC@5deg: 57.99 -> 57.21 | Hurt |
| replace NMS with SIFT post-processing (ablation) | CasMTR-4c · MegaDepth | AUC@5deg: 56.34 -> 51.84 | Hurt |
| NMS kernel 7x7 instead of 5x5 (ablation) | CasMTR-4c+2c · MegaDepth | AUC@5deg: 59.08 -> 57.64 | Hurt |
| apply NMS to CasMTR on HPatches (ablation) | CasMTR-4c · HPatches | AUC@3px: 67.5 -> 69.7 | Improved |
Closest papers by meaning
- LoFTR: Detector-Free Local Feature Matching With Transformers (CVPR 2021 · similarity 0.88)
- Semi-Dense Feature Matching With Transformers and its Applications in Multiple-View Geometry (TPAMI 2022 · similarity 0.86)
- Collaborative Feature Matching with Progressive Correspondence Learning (AAAI 2026 · similarity 0.86)
- Pose Recognition With Cascade Transformers (CVPR 2021 · similarity 0.86)
- ETO:Efficient Transformer-based Local Feature Matching by Organizing Multiple Homography Hypotheses (NeurIPS 2024 · similarity 0.86)
- Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed (CVPR 2024 · 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.
