NoiseSDF2NoiseSDF: Learning Clean Neural Fields from Noisy Supervision
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911183353675776 |
|---|---|
| author | Wang, Tengkai Li, Weihao Cui, Ruikai Qiu, Shi Barnes, Nick |
| author_facet | Wang, Tengkai Li, Weihao Cui, Ruikai Qiu, Shi Barnes, Nick |
| contents | Reconstructing accurate implicit surface representations from point clouds remains a challenging task, particularly when data is captured using low-quality scanning devices. These point clouds often contain substantial noise, leading to inaccurate surface reconstructions. Inspired by the Noise2Noise paradigm for 2D images, we introduce NoiseSDF2NoiseSDF, a novel method designed to extend this concept to 3D neural fields. Our approach enables learning clean neural SDFs directly from noisy point clouds through noisy supervision by minimizing the MSE loss between noisy SDF representations, allowing the network to implicitly denoise and refine surface estimations. We evaluate the effectiveness of NoiseSDF2NoiseSDF on benchmarks, including the ShapeNet, ABC, Famous, and Real datasets. Experimental results demonstrate that our framework significantly improves surface reconstruction quality from noisy inputs. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_13595 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | NoiseSDF2NoiseSDF: Learning Clean Neural Fields from Noisy Supervision Wang, Tengkai Li, Weihao Cui, Ruikai Qiu, Shi Barnes, Nick Computer Vision and Pattern Recognition Reconstructing accurate implicit surface representations from point clouds remains a challenging task, particularly when data is captured using low-quality scanning devices. These point clouds often contain substantial noise, leading to inaccurate surface reconstructions. Inspired by the Noise2Noise paradigm for 2D images, we introduce NoiseSDF2NoiseSDF, a novel method designed to extend this concept to 3D neural fields. Our approach enables learning clean neural SDFs directly from noisy point clouds through noisy supervision by minimizing the MSE loss between noisy SDF representations, allowing the network to implicitly denoise and refine surface estimations. We evaluate the effectiveness of NoiseSDF2NoiseSDF on benchmarks, including the ShapeNet, ABC, Famous, and Real datasets. Experimental results demonstrate that our framework significantly improves surface reconstruction quality from noisy inputs. |
| title | NoiseSDF2NoiseSDF: Learning Clean Neural Fields from Noisy Supervision |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2507.13595 |