Point Cloud Outlier Removal with Corruption-Stable Tokenization
Open Access
- 13 November 2026
- journal article
- research article
- Published by MDPI AG in Electronics
- Vol. 15 (22) , 5272
Abstract
Real-world point clouds often contain spurious observations that obscure the underlying geometry. Modern point-cloud models struggle to distinguish such outliers from sparse but valid local geometry. We propose a two-stage transformer framework for additive outlier removal that combines self-supervised masked reconstruction with supervised point-wise classification. To reduce pre-encoding corruption, the framework replaces data-dependent token centers with fixed spatial centers, drawing on Vision Transformers. This establishes a stable token layout while still allowing token contents to vary. The model retains points classified as structural without changing their coordinates. Experiments on normalized object-level point clouds with synthetic additive corruption show that the method removes a substantial proportion of outliers while preserving most structural points. Self-supervised pretraining is critical to this balance, while alternative tokenization changes the removal-preservation trade-off. Performance remains stable across tested contamination levels, although near-surface outliers and thin or sparsely sampled structures remain challenging.Keywords
This publication has 16 references indexed in Scilit:
- TripleMixer: A Triple-Domain Mixing Model for Point Cloud Denoising Under Adverse WeatherIEEE Transactions on Image Processing, 2025
- PointCVaR: Risk-Optimized Outlier Removal for Robust 3D Point Cloud ClassificationProceedings of the AAAI Conference on Artificial Intelligence, 2024
- NoiseTrans: Point Cloud Denoising with TransformersarXiv, 2023
- Masked Autoencoders for Point Cloud Self-supervised Learning2022
- SLiDE: Self-supervised LiDAR De-snowing Through Reconstruction Difficulty2022
- Using Shape to Categorize: Low-Shot Learning with an Explicit Shape Bias2021
- 3D Point Cloud Denoising Using Graph Laplacian Regularization of a Low Dimensional Manifold ModelIEEE Transactions on Image Processing, 2019
- PointCleanNet: Learning to Denoise and Remove Outliers from Dense Point CloudsComputer Graphics Forum, 2019
- PU-Net: Point Cloud Upsampling Network2018
- Point Cloud Denoising via Moving RPCAComputer Graphics Forum, 2016