Research map / Point clouds / Shape and surface completion
Noise2Score3D: Tweedie's Approach for Unsupervised Point Cloud Denoising
Xiangbin Wei, Yuanfeng Wang, Ao Xu, Lingyu Zhu, Dongyong Sun, Keren Li, Yang Li, Qi Qin
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 |
|---|---|---|---|
| use AR-DAE loss instead of DAE loss (ablation) | Noise2Score3D · ModelNet-40 50k, 1% noise | Chamfer distance x1e4: 5.153 -> 1.881 | Improved |
| use AR-DAE loss instead of DAE loss (ablation) | Noise2Score3D · ModelNet-40 50k, 3% noise | Chamfer distance x1e4: 34.999 -> 8.917 | Improved |
| use AR-DAE loss instead of DAE loss (ablation) | Noise2Score3D · ModelNet-40 50k, 2% noise | point-to-mesh x1e4: 16.092 -> 4.002 | Improved |
| anneal sigma_t during training instead of fixed 0.01 (ablation) | Noise2Score3D · ModelNet-40, 1% noise | Chamfer distance x1e4: 2.014 -> 1.881 | Improved |
| anneal sigma_t during training instead of fixed 0.03 (ablation) | Noise2Score3D · ModelNet-40, 3% noise | Chamfer distance x1e4: 8.511 -> 8.917 | Hurt |
| anneal sigma_t during training instead of fixed 0.01 (ablation) | Noise2Score3D · PU-Net, 1% noise | Chamfer distance x1e4: 1.254 -> 0.833 | Improved |
| anneal sigma_t during training instead of fixed 0.01 (ablation) | Noise2Score3D · PU-Net, 3% noise | Chamfer distance x1e4: 2.395 -> 2.434 | Hurt |
| use TV_PC-estimated sigma instead of true sigma (ablation) | Noise2Score3D · PU-Net 10K, 2% noise | Chamfer distance x1e4: 3.868 -> 4.190 | Hurt |
| use TV_PC-estimated sigma instead of true sigma (ablation) | Noise2Score3D · PU-Net 50K, 1% noise | Chamfer distance x1e4: 0.991 -> 0.833 | Improved |
| use KPConv backbone instead of PointNet++ (ablation) | Noise2Score3D · PU-Net 10K, 3% noise | Chamfer distance x1e4: 13.382 -> 5.583 | Improved |
| use KPConv backbone instead of original backbone (ablation) | DMR-U · PU-Net 10K, 1% noise | Chamfer distance x1e4: 5.313 -> 33.89 | Hurt |
| use KPConv backbone instead of original backbone (ablation) | Score-U · PU-Net 10K, 1% noise | Chamfer distance x1e4: 3.107 -> 3.494 | Hurt |
Later papers that compared with this method
1 later accepted paper reports Noise2Score3D as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- SIMPC: Learning Self-Induced Mirror-Point Consistency for Unsupervised Point Cloud Denoising (ICML 2026 · better on 11 of 11 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
- U-CAN: Unsupervised Point Cloud Denoising with Consistency-Aware Noise2Noise Matching (NeurIPS 2025 · similarity 0.92)
- You Should Learn to Stop Denoising on Point Clouds in Advance (AAAI 2025 · similarity 0.90)
- A Conditional Denoising Diffusion Probabilistic Model for Point Cloud Upsampling (CVPR 2024 · similarity 0.88)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · similarity 0.87)
- On the Posterior Distribution in Denoising: Application to Uncertainty Quantification (ICLR 2024 · similarity 0.87)
- Point Cloud Upsampling Using Conditional Diffusion Module with Adaptive Noise Suppression (CVPR 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.
