Research map / Point clouds / Shape and surface completion
Diffusion Bridges for 3D Point Cloud Denoising
Mathias Vogel Hüni, Keisuke Tateno, Marc Pollefeys, Federico Tombari, Marie-Julie Rakotosaona, Francis Engelmann
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 |
|---|---|---|---|
| align noisy and clean point sets with shortest-path assignment before training (ablation) | P2P-Bridge · PU-Net | CD (x1e4): 49.33 -> 2.11 | Improved |
| use deterministic OT-ODE rather than adding stochasticity to the bridge path (ablation) | P2P-Bridge · PU-Net | CD (x1e4): 2.45 -> 2.11 | Improved |
| use PVCNN backbone instead of Minkowski or SetTransformer (ablation) | P2P-Bridge · ScanNet++ subset | CD (x1e4): 12.79 -> 9.76 | Improved |
| add feature embedding layer (ablation) | P2P-Bridge · ScanNet++ subset | CD (x1e4): 9.76 -> 9.26 | Improved |
| double base channels from 32 to 64 (ablation) | P2P-Bridge · ScanNet++ subset | CD (x1e4): 9.26 -> 8.31 | Improved |
| add RGB features to XYZ input | P2P-Bridge · ARKitScenes | CD (x1e3): 20.14 -> 20.05 | Improved |
| add DINOv2 point features on top of RGB | P2P-Bridge · ARKitScenes | CD (x1e3): 20.05 -> 19.92 | Improved |
| add RGB and DINO features to XYZ input | P2P-Bridge · ScanNet++ (Apple LiDAR) | CD (x1e4): 64.46 -> 64.44 | No clear effect |
| P2P-Bridge vs PD-Flow at 1% Gaussian noise | P2P-Bridge · PU-Net (10k points, 1% noise) | Chamfer distance (x1e4): 2.13 -> 2.28 | Hurt |
| P2P-Bridge vs ScoreDenoise on real scans | P2P-Bridge · ScanNet++ (Apple LiDAR + 3DMatch) | CD (x1e4): 58.78 -> 35.56 | Improved |
Closest papers by meaning
- PointInfinity: Resolution-Invariant Point Diffusion Models (CVPR 2024 · similarity 0.88)
- TIGER: Time-Varying Denoising Model for 3D Point Cloud Generation with Diffusion Process (CVPR 2024 · similarity 0.88)
- SuperPC: A Single Diffusion Model for Point Cloud Completion, Upsampling, Denoising, and Colorization (CVPR 2025 · similarity 0.88)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.88)
- U-CAN: Unsupervised Point Cloud Denoising with Consistency-Aware Noise2Noise Matching (NeurIPS 2025 · similarity 0.88)
- A Conditional Denoising Diffusion Probabilistic Model for Point Cloud Upsampling (CVPR 2024 · 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.
