Risk-optimized Outlier Removal for Robust 3D Point Cloud Classification

TL;DR

The paper tackles the reliability of 3D point cloud classifiers under noisy and adversarial inputs by introducing PointCVaR, a tail-risk–driven outlier removal framework. It leverages gradient-based attribution to define per-point risk and formulates outlier removal as a CVaR minimization problem, solved efficiently via a linear programming relaxation to yield binary point weights. The approach demonstrates robust performance against random noise, adversarial intrusions, and backdoor triggers, and can be deployed as a plugin to strengthen existing robust classification pipelines. Empirically, PointCVaR improves accuracy on ModelNet40 and ShapeNet across multiple architectures, while providing interpretable, risk-based denoising with low computational overhead.

Abstract

With the growth of 3D sensing technology, deep learning system for 3D point clouds has become increasingly important, especially in applications like autonomous vehicles where safety is a primary concern. However, there are also growing concerns about the reliability of these systems when they encounter noisy point clouds, whether occurring naturally or introduced with malicious intent. This paper highlights the challenges of point cloud classification posed by various forms of noise, from simple background noise to malicious backdoor attacks that can intentionally skew model predictions. While there's an urgent need for optimized point cloud denoising, current point outlier removal approaches, an essential step for denoising, rely heavily on handcrafted strategies and are not adapted for higher-level tasks, such as classification. To address this issue, we introduce an innovative point outlier cleansing method that harnesses the power of downstream classification models. By employing gradient-based attribution analysis, we define a novel concept: point risk. Drawing inspiration from tail risk minimization in finance, we recast the outlier removal process as an optimization problem, named PointCVaR. Extensive experiments show that our proposed technique not only robustly filters diverse point cloud outliers but also consistently and significantly enhances existing robust methods for point cloud classification.

Figures (6)

• Figure 1: Visualizing point risks for point clouds with various types of outliers. It shows that the noise points often pose high risks compared to the clean points.
• Figure 2: Illustration of CVaR&VaR on risk distribution.
• Figure 3: The proposed framework of outlier removal by PointCVaR. Point risks are obtained by entering the noise sample into a trained classification model. Subsequently, an optimization problem is solved to minimize the tail of risk distribution, which leads to binary weights for noise point removal. The processed point cloud is utilized for classification.
• Figure 4: CVaR$_{0.99}$ boxplots (Mean, median and $25\backslash75$ quantiles) of clean sets risks (blue) and noisy sets risks (red).
• Figure 5: Accuracy on ModelNet40 without or with different noise vs. retention rate of PointCVaR.