Research map / 3D vision / Face and shape
TORA: Topological Representation Alignment for 3D Shape Assembly
Nahyuk Lee, Zhiang Chen, Marc Pollefeys, Sunghwan Hong
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 CKA topological alignment to RPF | RPF · Breaking Bad Everyday (21-33 parts) | Part accuracy (%): 62.1 -> 71.7 | Improved |
| add token-wise cosine alignment to RPF | RPF · Breaking Bad Everyday (2-20 parts) | Part accuracy (%): 93.2 -> 95.7 | Improved |
| add CKA alignment to RPF | RPF · PartNet-Assembly | Part accuracy (%): 59.8 -> 69.1 | Improved |
| add NT-Xent alignment to RPF (ablation) | RPF · TwoByTwo | Part accuracy (%): 65.4 -> 60.4 | Hurt |
| replace cosine alignment with CKA (ablation) | TORA · TwoByTwo | Part accuracy (%): 68.9 -> 71.5 | Improved |
| add cos-dist loss on top of CKA (ablation) | TORA (CKA) · TwoByTwo | Part accuracy (%): 71.5 -> 70.0 | Hurt |
| add NT-Xent loss on top of CKA (ablation) | TORA (CKA) · TwoByTwo | Part accuracy (%): 71.5 -> 68.5 | Hurt |
| add both cos-dist and NT-Xent on top of CKA (ablation) | TORA (CKA) · TwoByTwo | Part accuracy (%): 71.5 -> 67.7 | Hurt |
| add CKA alignment to RPF (zero-shot) | RPF · Zero-shot from Breaking Bad Everyday: FRACTURA | Part accuracy (%): 68.1 -> 76.0 | Improved |
| add NT-Xent alignment to RPF (zero-shot) (ablation) | RPF · Zero-shot from Breaking Bad Everyday: Breaking Bad Artifact | Part accuracy (%): 88.3 -> 87.0 | Hurt |
| use CKA instead of cosine with weak teacher PatchAlign3D (ablation) | TORA · TwoByTwo | Part accuracy (%): 66.0 -> 68.4 | Improved |
| align at deeper layer (l*=5) instead of early layers (ablation) | TORA · Breaking Bad | Part accuracy (%) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GARF — GARF: Learning Generalizable 3D Reassembly for Real-World Fractures (ICCV 2025) · better on 4 of 4 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
- Coarse-to-Fine 3D Part Assembly via Semantic Super-Parts and Symmetry-Aware Pose Estimation (NeurIPS 2025 · similarity 0.88)
- SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation (NeurIPS 2026 · similarity 0.88)
- Neural Shape Mating: Self-Supervised Object Assembly With Adversarial Shape Priors (CVPR 2022 · similarity 0.88)
- 3D Geometric Shape Assembly via Efficient Point Cloud Matching (ICML 2024 · similarity 0.87)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.86)
- Flow-Corrected Shape Optimization: Taming Manifold Drift in High-Dimensional 3D Models (NeurIPS 2026 · similarity 0.86)
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.
