Research map / 3D vision / Face and shape
Teaching DINOv3 About Partial 3D Geometry: A Self-Supervised Geometry-Aware Approach
Viktoria Ehm, Dongliang Cao, Riccardo Marin, Daniel Scholz, Weikang Wang, Florian Bernard, Daniel Cremers
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 |
|---|---|---|---|
| adapt DINOv3 with self-supervised GeoLoRA on synthetic partial shapes | DINOv3 features · BECOS partial-to-full (raw features) | geodesic error x100: 19.95 -> 5.57 | Improved |
| use geodesic-aware PointInfoNCE instead of vanilla PointInfoNCE (ablation) | GeoLoRA · BECOS validation | geodesic error x100: 11.29 -> 5.89 | Improved |
| use geodesic-aware loss instead of PointInfoNCE for chirality (ablation) | GeoLoRA · BECOS partial shapes chirality | left-right accuracy: 84.09 -> 91.42 | Improved |
| replace DINOv3 features with GeoLoRA features for chirality network | chirality network of [58] · BECOS partial shapes chirality | left-right accuracy: 70.45 -> 91.42 | Improved |
| swap DINOv3 features for GeoLoRA features in ULRSSM | ULRSSM · SHREC16 CUTS partial-to-full | geodesic error x100: 5.94 -> 3.01 | Improved |
| swap DINOv3 features for GeoLoRA features in DPFM | DPFM · BECOS partial-to-full | geodesic error x100: 23.16 -> 13.38 | Improved |
| swap DINOv3 features for GeoLoRA features in EchoMatch | EchoMatch · BECOS partial-to-partial | geodesic error x100: 9.74 -> 5.55 | Improved |
| swap DINOv3 features for GeoLoRA features in EchoMatch | EchoMatch · CP2P24 partial-to-partial | geodesic error x100: 2.97 -> 2.97 | No clear effect |
| swap DINOv3 features for GeoLoRA features in DPFM | DPFM · CP2P24 partial-to-partial | mIoU: 71.55 -> 70.41 (error 3.89 -> 3.48) | Mixed |
| increase number of views from 16 to 49 with frozen DINOv3 (ablation) | DINOv3 features · BECOS validation | geodesic error x100: 25.47 -> 20.09 | Improved |
| increase number of views from 49 to 100 (ablation) | DINOv3 features · BECOS validation | geodesic error x100: 20.09 -> 20.00 | No clear effect |
Closest papers by meaning
- Partial-to-Partial Shape Matching with Geometric Consistency (CVPR 2024 · similarity 0.88)
- Deep Confidence Guided Distance for 3D Partial Shape Registration (AAAI 2022 · similarity 0.86)
- EchoMatch: Partial-to-Partial Shape Matching via Correspondence Reflection (CVPR 2025 · similarity 0.85)
- DinoComplete: 3D Shape Completion with Distilled Semantic Priors and State Space Models (NeurIPS 2026 · similarity 0.85)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.85)
- Learning SO(3)-Invariant Semantic Correspondence via Local Shape Transform (CVPR 2024 · 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.
