Research map / 3D vision / Face and shape
Universal 3D Shape Matching via Coarse-to-Fine Language Guidance
Qinfeng Xiao, Guofeng Mei, Bo Yang, Liying Zhang, Jian Zhang, Kit-lun Yick
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 semantic feature fields (SD-DINO) to geometric features (ablation) | UniMatch · SNIS (inter-class) | average geodesic error: 0.49 -> 0.22 | Improved |
| Add group-wise rank-based contrastive loss (ablation) | UniMatch · SNIS (inter-class) | average geodesic error: 0.22 -> 0.19 | Improved |
| Add group-wise rank-based contrastive loss (ablation) | UniMatch · TOSCA (inter-class) | average geodesic error: 0.26 -> 0.23 | Improved |
| Use group-wise RnC loss instead of SupCon loss with pseudo positives (ablation) | UniMatch · TOSCA (inter-class) | average geodesic error: 0.29 -> 0.23 | Improved |
| Use FG-CLIP instead of CLIP language embeddings (ablation) | UniMatch · TOSCA (inter-class) | average geodesic error: 0.26 -> 0.23 | Improved |
| Use FG-CLIP instead of SigLip language embeddings (ablation) | UniMatch · TOSCA (inter-class) | average geodesic error: 0.24 -> 0.23 | No clear effect |
| Use FG-CLIP instead of CLIP on SHREC07 (ablation) | UniMatch · SHREC07 (inter-class) | average geodesic error: 0.37 -> 0.37 | No clear effect |
| Replace DenseMatcher with UniMatch on non-isometric SMAL | UniMatch · SMAL (non-isometric) | average geodesic error (x100): 4.7 -> 4.8 | No clear effect |
| Replace DenseMatcher with UniMatch on TOPKIDS | UniMatch · TOPKIDS (non-isometric) | average geodesic error (x100): 6.2 -> 5.9 | Improved |
| UniMatch versus URSSM on near-isometric FAUST | UniMatch · FAUST (near-isometric) | average geodesic error (x100): 1.6 -> 1.6 | No clear effect |
| UniMatch versus DenseMatcher on SHREC19 | UniMatch · SHREC19 (near-isometric) | average geodesic error (x100): 3.1 -> 3.2 | Hurt |
| Use language embeddings of MLLM part names, not hard correspondences | UniMatch · chair category (in-the-wild) | qualitative matching | Hurt |
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.
- DenseMatcher — DenseMatcher: Learning 3D Semantic Correspondence for Category-Level Manipulation from a Single Demo (ICLR 2025) · better on 5 of 7 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
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.89)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 2025 · similarity 0.89)
- UniDSeg: Unified Cross-Domain 3D Semantic Segmentation via Visual Foundation Models Prior (NeurIPS 2024 · similarity 0.87)
- Building a Strong Pre-Training Baseline for Universal 3D Large-Scale Perception (CVPR 2024 · similarity 0.87)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.87)
- DenseMatcher: Learning 3D Semantic Correspondence for Category-Level Manipulation from a Single Demo (ICLR 2025 · similarity 0.87)
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.
