Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
U-CAN: Unsupervised Point Cloud Denoising with Consistency-Aware Noise2Noise Matching
Authors: Junsheng Zhou, XingYu Shi, Haichuan Song, Yi Fang, Yu-Shen Liu, Zhizhong Han
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our evaluations under the widely used benchmarks in point cloud denoising, upsampling and image denoising show significant improvement over the state-of-the-art unsupervised methods, where U-CAN also produces comparable results with the supervised methods. Extensive experiments demonstrate that the proposed U-CAN outperforms state-of-the-art methods in unsupervised point cloud denoising, upsampling and image denoising, where U-CAN even achieves comparable performances with the supervised methods. Section 4, titled 'Experiments', details 'Point Cloud Denoising on Synthetic Data', 'Evaluatioins in Image Denoising', 'Point Upsampling via Denoising', and 'Ablation Studies', all of which involve empirical studies and data analysis. |
| Researcher Affiliation | Academia | School of Software, Tsinghua University, Beijing, China 1 Computer Science and Technology, East China Normal University, Shanghai, China 2 Center for AI and Robotics (CAIR), NYU Abu Dhabi, UAE 3 Department of Computer Science, Wayne State University, Detroit, USA4 |
| Pseudocode | No | The paper describes the method and architecture in text and with diagrams (Figure 1) but does not include a distinct pseudocode or algorithm block. |
| Open Source Code | Yes | We provide our demonstration code as a part of our supplementary materials. We will release the source code, data and instructions upon acceptance. |
| Open Datasets | Yes | For the experiments on synthetic shapes, we follow Score Denoise [27] to train our network on the PUNet [53] dataset. For evaluating in the image denoising task, we follow ZS-N2N [30] to conduct experiments under the Mc Master18 dataset [20]. We conduct evaluations under the Paris-rue-Madame dataset [43] which is obtained from real world using laser scanners. |
| Dataset Splits | Yes | For the experiments on synthetic shapes, we follow Score Denoise [27] to train our network on the PUNet [53] dataset. We split the dataset into training and testing sets with the same setting as Score Denoise [27]. |
| Hardware Specification | No | The computer resources needed to reproduce the experiments are provided in the appendix. (However, the appendix content with specific hardware details is not included in the provided paper text.) |
| Software Dependencies | No | The paper discusses various deep learning models and techniques, such as dynamic Edge Conv from DGCNN, but does not specify software dependencies with version numbers, such as Python or PyTorch versions, in the provided text. |
| Experiment Setup | No | The paper mentions parameters like patch size (1K), point cloud resolutions (10k, 50k), Gaussian noise levels (1%, 2%, 3%), Poisson noise levels (λ = 10, 25, 50), and the number of denoising steps (N=4) but does not provide specific deep learning training hyperparameters like learning rate, batch size, or optimizer settings in the main text. |