Research map / Point clouds / Shape and surface completion
Total Denoising: Unsupervised Learning of 3D Point Cloud Cleaning
Pedro Hermosilla, Tobias Ritschel, Timo Ropinski
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 spatial proximity prior to unsupervised training (ablation) | Total Denoising · Simple noise, 22M points | Chamfer error: .582 -> .547 | Improved |
| add color (appearance) to the prior sampling (ablation) | Total Denoising · Simple noise, 22M points | Chamfer error: .547 -> .542 | Improved |
| add spatial prior (ablation) | Total Denoising · Advanced noise | Chamfer error: .393 -> .359 | Improved |
| add color to prior (ablation) | Total Denoising · Advanced noise | Chamfer error: .359 -> .356 | Improved |
| unsupervised Full instead of supervised training | Total Denoising · Simple noise, 22M points | Chamfer error: .545 -> .542 | Improved |
| unsupervised Full instead of supervised training | Total Denoising · Advanced noise | Chamfer error: .356 vs .329 | Hurt |
| use unsupervised training with little data (0.5M points) instead of supervised | Total Denoising · Simple noise, 0.5M points | Chamfer error: .574 -> .558 | Improved |
| increase training data from 0.5M to 22M points (ablation) | Total Denoising Full · Simple noise | Chamfer error: .558 -> .542 | Improved |
| use unsupervised Full instead of supervised at 1% noise | Total Denoising · Simple noise 1.0% | Chamfer error: .525 vs .515 | Hurt |
| use Gaussian kernel instead of Wendland or inverse multi-quadric for the prior (ablation) | Total Denoising · Prior kernel ablation | Chamfer error: .548 vs .560 vs .577 | Improved |
| train on a single noise level instead of all levels (ablation) | Total Denoising · Test 1.0% noise | Chamfer error: .534 -> .551 | Hurt |
| apply our unsupervised prior to PointCleanNet training | PCNet · PointCleanNet dataset | error: 1.36 -> 1.34 | No clear effect |
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 0.91)
- 3DMambaIPF: A State Space Model for Iterative Point Cloud Filtering via Differentiable Rendering (AAAI 2025 · similarity 0.88)
- Denoising Point Clouds in Latent Space via Graph Convolution and Invertible Neural Network (CVPR 2024 · similarity 0.88)
- TIGER: Time-Varying Denoising Model for 3D Point Cloud Generation with Diffusion Process (CVPR 2024 · similarity 0.88)
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud Classification (AAAI 2024 · similarity 0.88)
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.
