3D anomaly detection improves by modeling geometric relational consistency
Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling
Computer Vision and Pattern Recognition
Summary
Finding defects in 3D scanned objects is hard because normal shapes can vary a lot. The authors designed a method that looks for places where the geometric relationships between parts break down, which often signals a defect. Instead of just learning what normal 3D shapes look like, their approach learns what kinds of inconsistencies hint at problems. It works well even when tested on different categories or datasets, making it more reliable in real-world inspection tasks.
What this means in practice
- •For industrial inspection teams: Identify defective regions in 3D scanned industrial parts with improved accuracy and fewer false alarms.
- •For manufacturing quality control: Automate detection of subtle structural defects across various product categories in manufacturing workflows.
Authors
KunHo Heo, SuYeon Kim, Hayoung Lee, Chanse Oh, MyeongAh Cho
Abstract
3D anomaly detection (3DAD) aims to identify defective regions in point cloud data, serving as a critical component in industrial inspection systems. Existing methods are normality-centered -- learning the distribution of normal samples and treating deviations as anomalies -- without explicitly modeling what constitutes a defect. This leads to ambiguous decision boundaries with increased false positives and negatives, particularly in unified and cross-domain settings where diverse normal distributions further blur the boundaries. We propose a relational inconsistency modeling framework that characterizes defects as violations of geometric consistency among neighboring structures. Our approach learns category-agnostic defect cues through pseudo-anomalies designed as controlled relational violations, instantiated by two key modules: Edge-aware Graph Refinement (EGR) for encoding geometric relationships among local regions, and Cluster-Deviation Modeling (CDM) for identifying regions that are relationally incompatible within their structural peer group. Extensive experiments on Anomaly-ShapeNet and Real3D-AD demonstrate consistent improvements over prior state-of-the-art methods in both in-domain and cross-domain settings, validating the effectiveness of learning an explicit, relation-based defect criterion for 3D anomaly detection. Project page: https://visualsciencelab-khu.github.io/GRIM_project/.