CASL: Curvature-Augmented Self-supervised Learning for 3D Anomaly Detection

TL;DR

CASL addresses the generalization gap in 3D anomaly detection by leveraging curvature as a robust intrinsic cue within a reconstruction-based self-supervised framework. It introduces curvature-Augmented Self-supervised Learning (CASL), a multi-scale curvature-prompted, full-coordinate-masking architecture built on a Minkowski U-Net, and trains on normal data with a pseudo-anomaly fine-tuning stage for detection and localization. The method achieves state-of-the-art anomaly detection on Real3D-AD and Anomaly-ShapeNet, while also demonstrating strong transfer to standard 3D understanding tasks such as classification and part segmentation, despite pretraining on a relatively small normal-sample set. These results highlight curvature as a powerful geometric prior that mitigates geometric shortcuts and promotes deeper 3D shape understanding, with practical impact for robust industrial inspection and beyond.

Abstract

Deep learning-based 3D anomaly detection methods have demonstrated significant potential in industrial manufacturing. However, many approaches are specifically designed for anomaly detection tasks, which limits their generalizability to other 3D understanding tasks. In contrast, self-supervised point cloud models aim for general-purpose representation learning, yet our investigation reveals that these classical models are suboptimal at anomaly detection under the unified fine-tuning paradigm. This motivates us to develop a more generalizable 3D model that can effectively detect anomalies without relying on task-specific designs. Interestingly, we find that using only the curvature of each point as its anomaly score already outperforms several classical self-supervised and dedicated anomaly detection models, highlighting the critical role of curvature in 3D anomaly detection. In this paper, we propose a Curvature-Augmented Self-supervised Learning (CASL) framework based on a reconstruction paradigm. Built upon the classical U-Net architecture, our approach introduces multi-scale curvature prompts to guide the decoder in predicting the spatial coordinates of each point. Without relying on any dedicated anomaly detection mechanisms, it achieves leading detection performance through straightforward anomaly classification fine-tuning. Moreover, the learned representations generalize well to standard 3D understanding tasks such as point cloud classification. The code is available at https://github.com/zyh16143998882/CASL.

Figures (8)

• Figure 1: Anomaly detection performance of different methods on the Real3D-AD dataset. Blue markers represent task-specific anomaly detection methods, while green markers correspond to classical self-supervised models fine-tuned under the “pre-training and fine-tuning” paradigm. Red markers denote our results: Curvature refers to our non-learning-based method, and CASL represents the proposed curvature-augmented self-supervised learning framework.
• Figure 2: (a) Curvature heatmap of an anomalous point cloud, where warmer colors indicate higher curvature. (b) Spatial distribution of normal (green) and abnormal (red) points. Anomalous point boundaries often exhibit sharp increases in curvature.
• Figure 3: The pipeline of our Curvature-Augmented Self-supervised Learning (CASL) framework. Our framework, built upon the U-Net architecture, leverages multi-scale curvature prompts to guide the recovery of spatial coordinates for all masked points from random initialization.
• Figure 4: Detection performance with varying mask ratios.
• Figure 5: Detailed structure of the encoder block.