Research map / Detection and segmentation / CNN features
CasP: Improving Semi-Dense Feature Matching Pipeline Leveraging Cascaded Correspondence Priors for Guidance
Peiqi Chen, Lei Yu, Yi Wan, Yingying Pei, Xinyi Liu, Yongxiang Yao, Yingying Zhang, Lixiang Ru, Liheng Zhong, Jingdong Chen, Ming Yang, Yongjun Zhang
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 dual-softmax back to ELoFTR matching (ablation) | ELoFTR · ETH3D[O]-3438 (RANSAC) | AUC@5deg: 53.4 -> 56.7 | Improved |
| add dual-softmax back to ELoFTR matching (ablation) | ELoFTR · ETH3D[I]-2131 (RANSAC) | AUC@5deg: 44.7 -> 49.1 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (full) (ablation) | ELoFTR · ETH3D[O]-3438 (RANSAC) | AUC@5deg: 56.7 -> 60.1 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (full) (ablation) | ELoFTR · ETH3D[I]-2131 (RANSAC) | AUC@5deg: 49.1 -> 52.3 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (lite) (ablation) | ELoFTR · ETH3D[I]-2131 (RANSAC) | AUC@10deg and AUC@20deg: 55.0 -> 54.9 (@10), 59.4 -> 58.6 (@20) | Mixed |
| add the other advanced modules (hybrid interaction, RSCA, homography refinement) on top of cascaded matching (ablation) | EL+CM-full · ETH3D[O]-3438 (RANSAC) | AUC@5deg: 60.1 -> 61.8 | Improved |
| add the other advanced modules on top of cascaded matching (ablation) | EL+CM-full · ETH3D[I]-2131 (RANSAC) | AUC@5deg: 52.3 -> 56.1 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (full), use MAGSAC++ (ablation) | ELoFTR · ETH3D[O]-3438 (MAGSAC++) | AUC@5deg: 58.2 -> 61.8 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (full) (ablation) | ELoFTR · MD-1500 efficiency | runtime FP32 (ms): 238.3 -> 144.6 | Improved |
| remove dual-softmax from ELoFTR (ablation) | ELoFTR · MD-1500 efficiency | runtime FP32 (ms): 238.3 -> 185.3 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (full) (ablation) | ELoFTR · MD-1500 efficiency | GMACs: 909.1 -> 708.1 | Improved |
| replace dual-softmax in ELoFTR with cascaded matching (full) (ablation) | ELoFTR · MD-1500 efficiency | memory FP16 (GB): 13.6 -> 7.4 | Improved |
Closest papers by meaning
- HomoMatcher: Achieving Dense Feature Matching with Semi-Dense Efficiency by Homography Estimation (AAAI 2025 · similarity 0.87)
- RoMa v2: Harder Better Faster Denser Feature Matching (ECCV 2026 · similarity 0.86)
- Improving Transformer-based Image Matching by Cascaded Capturing Spatially Informative Keypoints (ICCV 2023 · similarity 0.86)
- RoMa: Robust Dense Feature Matching (CVPR 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)
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.
