PointCore: Efficient Unsupervised Point Cloud Anomaly Detector Using Local-Global Features

TL;DR

This work proposes an unsupervised point cloud anomaly detection framework based on joint local-global features, termed PointCore, which achieves competitive inference time and the best performance in both detection and localization as compared to the state-of-the-art Reg3D-AD approach and several competitors.

Abstract

Three-dimensional point cloud anomaly detection that aims to detect anomaly data points from a training set serves as the foundation for a variety of applications, including industrial inspection and autonomous driving. However, existing point cloud anomaly detection methods often incorporate multiple feature memory banks to fully preserve local and global representations, which comes at the high cost of computational complexity and mismatches between features. To address that, we propose an unsupervised point cloud anomaly detection framework based on joint local-global features, termed PointCore. To be specific, PointCore only requires a single memory bank to store local (coordinate) and global (PointMAE) representations and different priorities are assigned to these local-global features, thereby reducing the computational cost and mismatching disturbance in inference. Furthermore, to robust against the outliers, a normalization ranking method is introduced to not only adjust values of different scales to a notionally common scale, but also transform densely-distributed data into a uniform distribution. Extensive experiments on Real3D-AD dataset demonstrate that PointCore achieves competitive inference time and the best performance in both detection and localization as compared to the state-of-the-art Reg3D-AD approach and several competitors.

Figures (4)

• Figure 1: Heatmaps of anomaly scores obtained by several methods on the Real3D-AD dataset. From the visualization, we see that the proposed method can detect and locate anomaly data points more accurately compared to others.
• Figure 2: The pipeline of PointCore architecture. We randomly select and convert template point clouds as the reference coordinates through global registration and local optimization methods, use pre-trained PointMAE feature extractor to obtain the PointMAE features, and then bind the coordinates and the features to establish the coordinate-PointMAE memory bank. Finally, we compute anomaly scores for all points in inference.
• Figure 3: The process of the PointMAE feature interpolation.
• Figure 4: Distribution of coordinate and PointMAE anomaly scores normalized by these two methods.