A Lightweight 3D Anomaly Detection Method with Rotationally Invariant Features

3D anomaly detection (AD) is a crucial task in computer vision, aiming to identify anomalous points or regions from point cloud data. However, existing methods may encounter challenges when handling point clouds with changes in orientation and position because the resulting features may vary significantly. To address this problem, we propose a novel Rotationally Invariant Features (RIF) framework for 3D AD. Firstly, to remove the adverse effect of variations on point cloud data, we develop a Point Coordinate Mapping (PCM) technique, which maps each point into a rotationally invariant space to maintain consistency of representation. Then, to learn robust and discriminative features, we design a lightweight Convolutional Transform Feature Network (CTF-Net) to extract rotationally invariant features for the memory bank. To improve the ability of the feature extractor, we introduce the idea of transfer learning to pre-train the feature extractor with 3D data augmentation. Experimental results show that the proposed method achieves the advanced performance on the Anomaly-ShapeNet dataset, with an average P-AUROC improvement of 17.7%, and also gains the best performance on the Real3D-AD dataset, with an average P-AUROC improvement of 1.6%. The strong generalization ability of RIF has been verified by combining it with traditional feature extraction methods on anomaly detection tasks, demonstrating great potential for industrial applications.

💡 Research Summary

The paper addresses a fundamental challenge in 3‑dimensional anomaly detection (3D‑AD) on point clouds: the sensitivity of learned features to arbitrary rotations and translations of the input data. Existing approaches either rely on costly registration steps, multi‑view synthesis, or data augmentation that only partially mitigates this issue. The authors propose a novel Rotationally Invariant Features (RIF) framework that eliminates the need for explicit alignment while remaining lightweight and computationally efficient.

The first component, Point Coordinate Mapping (PCM), transforms each raw point cloud into a rotation‑invariant coordinate system. For a given cloud, the centroid is computed, then three anchor points are selected: the farthest point from the centroid (p_far1), the second farthest point that is linearly independent of the first (p_far2), and the nearest point that ensures full rank of the three vectors. These vectors are orthogonalized via Gram‑Schmidt to obtain an orthonormal basis (e1, e2, e3). Every point p_i is then re‑expressed as p_i* = (p_i – c)·