Hypergraph Normal World Models for Logical Visual Anomaly Detection

2026-06-23

AI Summary

This work addresses the challenge of detecting logical visual anomalies—images that appear locally normal but violate global constraints such as object counts, co-occurrence patterns, or spatial relationships. The authors propose the first hypergraph-based framework for logical anomaly detection, which constructs a category-specific “normal world” model using only normal training images. By freezing DINOv2 to extract patch tokens, the method builds spatial hyperedges to explicitly model local, pairwise, and higher-order interactions among image regions. Multi-level statistical features are integrated and evaluated via information-theoretic scoring to produce a comprehensive anomaly score. On the MVTec LOCO breakfast-box dataset, the approach achieves an AUROC of 0.9279, substantially outperforming both the DINOv2 patch-kNN baseline (0.8434) and a non-hypergraph variant (0.9013), while maintaining strong performance under few-shot conditions.

Abstract

Visual anomaly detection is often deployed with only normal training images. Most one-class detectors map test patches or features to a normal reference distribution. This works well for local structural defects. Logical anomalies are different. Each visible part may look normal, while the whole image violates a normal count, co-occurrence, or spatial relation. This paper studies whether a model can learn such a category-specific normal world from nominal images alone. We propose the Hypergraph Normal World Model, a normal-only detector that distills frozen DINOv2 patch tokens into patch, relation, and hypergraph statistics. It builds spatial hyperedges over token groups. It then scores each test image with an information quotient that separates local, relational, hyperedge, and hyperedge-relation evidence. On the available MVTec LOCO breakfast-box validation data, the full hypergraph model improves logical anomaly AUROC from 0.8434 for DINOv2 patch-kNN to 0.9279. It also improves over the non-hypergraph variant, from 0.9013 to 0.9279. Few-shot experiments show that the model remains effective with very limited normal images. We also test whether the score reflects normal-world knowledge rather than a shallow mapping. t-SNE separates logical anomalies in the learned energy space. Relation counterfactuals increase the information quotient by 83.13 on average. Random hypergraphs reduce logical AUROC, and hyperedge attribution is much larger on logical anomalies. Qualitative examples show that high scores are driven by relation-bearing terms. These results suggest that logical visual anomaly detection should model normal relations, not only normal local patches.

Problem

• logical anomaly
• visual anomaly detection
• normal world model
• hypergraph
• relation modeling

Innovation

• hypergraph
• logical anomaly detection
• normal world model
• relation modeling
• visual anomaly detection