Unveiling Multi-View Anomaly Detection: Intra-view Decoupling and Inter-view Fusion

Kai Mao, Yiyang Lian, Yangyang Wang, Meiqin Liu, Nanning Zheng, Ping Wei

Abstract: Anomaly detection has garnered significant attention due to its extensive industrial application value. Most existing methods focus solely on single-view scenarios. They fail to detect anomalies hidden in blind spots, leaving a gap in addressing the demands of multi-view detection in practical applications. Ensemble of multiple single-view models is a typical way to tackle the multi-view situation, but it overlooks the correlations between different views. Based on the idea of Intra-view Decoupling and Inter-view Fusion, we propose a new multi-view anomaly detection framework called IDIF to exploit correlations among views. Our method contains three key components: (1) a proposed Consistency Bottleneck module which extracts common feature of different views through information compression and mutual information maximization; (2) an Implicit Voxel Construction module which fuses features of different views with prior knowledge represented in the form of voxels; and (3) a View-wise Dropout training strategy that allows the model to learn how to cope with missing views during testing. The proposed IDIF achieves 97.0% S-AUROC detection performance on the Real-IAD dataset, outperforming existing state-of-the-art methods by 2.1%. The method also achivevs state-of-the-art performance on two 3D dataset, with 95.6%, 94.2% S-AUROC on the MVTec 3D-AD and Eyecandies dataset respectively.