Research map / Point clouds / Shape and surface completion
SIMPC: Learning Self-Induced Mirror-Point Consistency for Unsupervised Point Cloud Denoising
Chengwei Zhang, Xueyi Zhang, Tao Jiang, Xinhao Xu, Wenjie Li, Fubo Zhang, Longyong Chen
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 mirror-point consistency loss on top of CD similarity loss (ablation) | SIMPC · PUNet 10K 3% Gaussian | Chamfer distance: 65.73 -> 34.42 | Improved |
| add mirror-point consistency loss on top of CD similarity loss (ablation) | SIMPC · PUNet 10K 3% Gaussian | P2M distance: 36.20 -> 13.85 | Improved |
| use mirror-point consistency instead of EMD similarity (ablation) | SIMPC · PUNet 10K 3% Gaussian | Chamfer distance: 41.04 -> 34.42 | Improved |
| use EMD similarity instead of CD only (ablation) | SIMPC · PUNet 10K 3% Gaussian | Chamfer distance: 65.73 -> 41.04 | Improved |
| use EMD instead of CD-only similarity loss (ablation) | SIMPC · PUNet 10K 1% Gaussian | Chamfer distance: 18.91 -> 26.54 | Hurt |
| use symmetric extension w2 = 2 instead of 1.5 (ablation) | SIMPC mirror-point generation · PUNet 10K 3% Gaussian | Chamfer distance: 35.10 -> 34.42 | Improved |
| use symmetric extension w2 = 2 instead of 2.5 (ablation) | SIMPC mirror-point generation · PUNet 10K 3% Gaussian | Chamfer distance: 36.33 -> 34.42 | Improved |
| SIMPC vs best prior unsupervised method U-CAN | SIMPC · PUNet 10K 1% Gaussian | Chamfer distance: 24.97 -> 20.25 | Improved |
| SIMPC vs Noise2Score3D on non-Gaussian noise | SIMPC · PUNet 10K 1% Laplacian | Chamfer distance: 43.08 -> 23.65 | Improved |
| SIMPC vs Score-U on real scans | SIMPC · Kinect | Chamfer distance: 15.85 -> 13.01 | Improved |
| SIMPC vs supervised PD-Refiner | SIMPC · PUNet 10K 1% Gaussian | Chamfer distance: 17.54 vs 20.25 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Noise2Score3D — Noise2Score3D: Tweedie's Approach for Unsupervised Point Cloud Denoising (ICCV 2025) · better on 11 of 11 shared results
- U-CAN — U-CAN: Unsupervised Point Cloud Denoising with Consistency-Aware Noise2Noise Matching (NeurIPS 2025) · better on 6 of 6 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
- Total Denoising: Unsupervised Learning of 3D Point Cloud Cleaning (ICCV 2019 · similarity 0.89)
- U-CAN: Unsupervised Point Cloud Denoising with Consistency-Aware Noise2Noise Matching (NeurIPS 2025 · similarity 0.89)
- PointInfinity: Resolution-Invariant Point Diffusion Models (CVPR 2024 · similarity 0.86)
- Where Precision Meets Efficiency: Transformation Diffusion Model for Point Cloud Registration (AAAI 2025 · similarity 0.86)
- You Should Learn to Stop Denoising on Point Clouds in Advance (AAAI 2025 · similarity 0.86)
- PLATYPUS: Progressive Local Surface Estimator for Arbitrary-Scale Point Cloud Upsampling (AAAI 2025 · similarity 0.86)
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.
