Research map / 3D vision / Depth and stereo
MV-RoMa: From Pairwise Matching into Multi-View Track Reconstruction
Jongmin Lee, Seungyeop Kang, Sungjoo Yoo
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 multi-view encoder (track-guided DINOv2) (ablation) | RoMa · ETH3D triangulation | accuracy@1cm / completeness@1cm: 75.58 -> 77.80 acc; 5.64 -> 6.75 comp | Improved |
| add multi-view refiner on top of encoder (ablation) | MV-RoMa with MV-Encoder · ETH3D triangulation | accuracy@1cm / completeness@1cm: 77.80 -> 81.91 acc; 6.75 -> 7.52 comp | Improved |
| use 1 track token instead of 512 (ablation) | MV-RoMa · HPatches (DLT) | AUC@1px: 41.8 -> 36.1 | Hurt |
| reduce track tokens from 512 to 16 (ablation) | MV-RoMa · HPatches (DLT) | AUC@1px: 41.8 -> 41.6 | No clear effect |
| process 5 views jointly instead of pairwise (ablation) | MV-RoMa · HPatches (DLT) | AUC@1px: 41.2 -> 41.8 | Improved |
| multi-view matching vs RoMa | MV-RoMa · HPatches (DLT) | AUC@1px: 41.0 -> 46.1 | Improved |
| multi-view matching vs RoMa | MV-RoMa · ETH3D triangulation (two-view tracks included) | accuracy@1cm: 75.58 -> 81.91 | Improved |
| compare to full SfM pipeline RoMa + Dense-SfM (two-view tracks excluded) | MV-RoMa · ETH3D triangulation | completeness@1cm: 7.38 -> 3.95 | Hurt |
| multi-view matching vs RoMa + Dense-SfM | MV-RoMa · IMC PhotoTourism pose | AUC@3deg: 48.48 -> 51.31 | Improved |
Closest papers by meaning
- CoMatcher: Multi-View Collaborative Feature Matching (CVPR 2025 · similarity 0.89)
- RoMa: Robust Dense Feature Matching (CVPR 2024 · similarity 0.89)
- Dense-SfM: Structure from Motion with Dense Consistent Matching (CVPR 2025 · similarity 0.87)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 2025 · similarity 0.85)
- HomoMatcher: Achieving Dense Feature Matching with Semi-Dense Efficiency by Homography Estimation (AAAI 2025 · similarity 0.85)
- Understanding Multi-View Transformers (NeurIPS 2026 · similarity 0.85)
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.
