Research map / Point clouds / Shape and surface completion
Fast Learning of Signed Distance Functions From Noisy Point Clouds via Noise to Noise Mapping
Junsheng Zhou, Baorui Ma, Yu-Shen Liu, Zhizhong Han
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 EMD instead of Chamfer distance as the noise-to-noise distance (ablation) | Noise2NoiseMapping · PU denoising | L2CD x1e4: 73.786 -> 4.221 | Improved |
| add geometric consistency regularization (lambda 0.1) to EMD for reconstruction (ablation) | Noise2NoiseMapping · PU reconstruction | L2CD x1e4: 80.917 -> 4.277 | Improved |
| increase regularization weight lambda from 0.1 to 0.2 (ablation) | Noise2NoiseMapping · PU reconstruction | L2CD x1e4: 4.277 -> 4.993 | Hurt |
| add regularization to denoising (lambda 0 vs 0.2) (ablation) | Noise2NoiseMapping · PU denoising | L2CD x1e4: 4.221 -> 4.832 | Hurt |
| increase number of noisy observations N from 1 to 200 (ablation) | Noise2NoiseMapping · PU denoising | L2CD x1e4: 4.976 -> 4.221 | Improved |
| increase batch size B from 250 to 10000 (ablation) | Noise2NoiseMapping · PU denoising | L2CD x1e4: 4.221 -> 6.217 | Hurt |
| train for more iterations (40e4 to 80e4) (ablation) | Noise2NoiseMapping · PU denoising | L2CD x1e4: 4.887 -> 4.221 | Improved |
| mix multiple noisy observations into one (ablation) | Noise2NoiseMapping · PU denoising | L2CD x1e4: 4.221 -> 4.244 | No clear effect |
| add pull loss to hash-grid fast learning (ablation) | Noise2NoiseMapping (+Fast) · PU denoising 10K | L2CD x1e4: 8.016 -> 4.537 | Improved |
| add Eikonal regularization to hash-grid fast learning (ablation) | Noise2NoiseMapping (+Fast) · PU denoising 10K | L2CD x1e4: 5.255 -> 4.537 | Improved |
| use multi-resolution hash encoding (Fast) | Noise2NoiseMapping · 3D Scene | training time: 913 s -> 63 s | Improved |
| use learned SDF prior from noisy SfM points to guide NeuS | NeuS · DTU (15 scenes) | Chamfer distance: 0.87 -> 0.72 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- POCO — POCO: Point Convolution for Surface Reconstruction (CVPR 2022) · better on 5 of 5 shared results
- DiGS — DiGS: Divergence Guided Shape Implicit Neural Representation for Unoriented Point Clouds (CVPR 2022) · better on 1 of 2 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
- Inferring Neural Signed Distance Functions by Overfitting on Single Noisy Point Clouds through Finetuning Data-Driven based Priors (NeurIPS 2024 · similarity 0.92)
- Toward Robust Neural Reconstruction from Sparse Point Sets (CVPR 2025 · similarity 0.89)
- MultiPull: Detailing Signed Distance Functions by Pulling Multi-Level Queries at Multi-Step (NeurIPS 2024 · similarity 0.88)
- 3DMambaIPF: A State Space Model for Iterative Point Cloud Filtering via Differentiable Rendering (AAAI 2025 · similarity 0.88)
- U-CAN: Unsupervised Point Cloud Denoising with Consistency-Aware Noise2Noise Matching (NeurIPS 2025 · similarity 0.88)
- A Lightweight UDF Learning Framework for 3D Reconstruction Based on Local Shape Functions (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.
