Research map / 3D vision / Depth and stereo
Guiding Local Feature Matching with Surface Curvature
Shuzhe Wang, Juho Kannala, Marc Pollefeys, Daniel Barath
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 curvature similarity extractor (CSE) with fine-tuning to QuadTree | QuadTree · ScanNet | AUC@5: 24.9 -> 25.8 | Improved |
| add CSE to SuperGlue without fine-tuning | SuperGlue · MegaDepth-1500 | AUC@5: 43.7 -> 46.2 | Improved |
| add CSE to SP+SG for visual localization | SuperPoint + SuperGlue · Aachen Day-Night Night (0.25m, 2deg) | pose recall (%): 77.0 -> 75.9 | Hurt |
| add CSE to SP+SG for visual localization | SuperPoint + SuperGlue · Aachen Day-Night Day (0.25m, 2deg) | pose recall (%): 89.8 -> 90.7 | Improved |
| fine-tune depth predictor with matching loss on top of CSE (ablation) | QuadTree + CSE · ScanNet | AUC@5: 25.2 -> 25.8 | Improved |
| use multi-scale curvature extraction (ablation) | QuadTree + CSE · ScanNet | AUC@5: 25.2 -> 25.6; time 112 -> 338 ms | Mixed |
| replace ViT MiDaS depth predictor with lightweight MiDaS (Real-time) (ablation) | QuadTree + CSE · ScanNet | AUC@5: 25.8 -> 25.5 | No clear effect |
| use ground-truth depth instead of fine-tuned predicted depth (ablation) | QuadTree + CSE · ScanNet | AUC@5: 25.8 -> 25.1 | Hurt |
| replace quadric (ellipse) fitting with surface fitting for curvature (ablation) | QuadTree + CSE · ScanNet | AUC@5: 25.8 -> 24.8 | Hurt |
| set mixing weight lambda lower than 0.8-0.9 (ablation) | QuadTree + CSE · ScanNet | AUC@5: 26.1 -> 17.8 (lambda 0.8 to 0.3) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- SuperGlue — SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020) · better on 3 of 3 shared results
- SP + SG — SPSG: Self-Supervised Photometric Scene Generation From RGB-D Scans (CVPR 2021) · better on 1 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
- 3DMatch: Learning Local Geometric Descriptors From RGB-D Reconstructions (CVPR 2017 · similarity 0.87)
- Wormhole Loss for Partial Shape Matching (NeurIPS 2024 · similarity 0.85)
- SegMASt3R: Geometry Grounded Segment Matching (NeurIPS 2025 · similarity 0.84)
- Improving Neural Network Surface Processing with Principal Curvatures (NeurIPS 2024 · similarity 0.84)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.83)
- CoMatcher: Multi-View Collaborative Feature Matching (CVPR 2025 · 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.
