Research map / 3D vision / Face and shape
Unsupervised Deep Learning for Structured Shape Matching
Jean-Michel Roufosse, Abhishek Sharma, Maks Ovsjanikov
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 |
|---|---|---|---|
| combine all four penalties instead of E3 alone (ablation) | SURFMNet · FAUST | mean geodesic error: 0.073 -> 0.020 | Improved |
| add E1 and E2 to E3 and E4 (vs E1,E3,E4 etc.) full set vs without E2 (ablation) | SURFMNet · FAUST | mean geodesic error: 0.077 -> 0.020 | Improved |
| add E3 (Laplacian commutativity) to E1+E2+E4 (ablation) | SURFMNet · FAUST | mean geodesic error: 0.126 -> 0.020 | Improved |
| add E4 to E1+E2+E3 (ablation) | SURFMNet · FAUST | mean geodesic error: 0.081 -> 0.020 | Improved |
| add E1 to E2+E3+E4 (ablation) | SURFMNet · FAUST | mean geodesic error: 0.079 -> 0.020 | Improved |
| use E3 alone instead of E4 alone (ablation) | SURFMNet · FAUST | mean geodesic error: 0.252 -> 0.073 | Improved |
| learn descriptors with the network instead of fixed SHOT descriptors (Fmap Ours Opt vs Fmap Basic) (ablation) | Fmap Basic · FAUST remeshed 5k | mean geodesic error (x1e-3): 366.2 -> 20.19 | Improved |
| add ICP post-processing | SURFMNet-sub · FAUST remeshed 5k | mean geodesic error (x1e-3): 92.09 -> 30.47 | Improved |
| add ICP post-processing | SURFMNet-sub · SCAPE remeshed 5k | mean geodesic error (x1e-3): 60.32 -> 23.00 | Improved |
| add PMF post-processing | FMNet · FAUST remeshed 5k | mean geodesic error (x1e-3): 112.8 -> 83.61 | Improved |
| optimize on all shapes (Ours-all) instead of the 80-shape training subset | SURFMNet · FAUST | mean geodesic error: 0.020 -> 0.008 | Improved |
| use unsupervised SURFMNet instead of supervised FMNet | SURFMNet · FAUST remeshed 5k | mean geodesic error (x1e-3): 112.8 -> 92.09 | Improved |
Closest papers by meaning
- Hyper-Network Neural Functional Maps for Unsupervised Robust 3D Shape Matching (ECCV 2026 · similarity 0.92)
- Deep Geometric Functional Maps: Robust Feature Learning for Shape Correspondence (CVPR 2020 · similarity 0.91)
- Unsupervised Contrastive Learning for Efficient and Robust Spectral Shape Matching (AAAI 2026 · similarity 0.91)
- NCP: Neural Correspondence Prior for Effective Unsupervised Shape Matching (NeurIPS 2022 · similarity 0.91)
- Integrating Efficient Optimal Transport and Functional Maps For Unsupervised Shape Correspondence Learning (CVPR 2024 · similarity 0.89)
- Spectral Meets Spatial: Harmonising 3D Shape Matching and Interpolation (CVPR 2024 · similarity 0.88)
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.
