IMMoE: Incomplete Multi-View Anomaly Detection via Mixture of View Experts Fusion

Existing Multi-view Anomaly Detection (MAD) methods as-sume that all views are completely available and model each view sep-arately. However, in real industrial scenarios, information in the viewmay be missing due to faults such as occlusion, which leads to the per-formance degradation of existing methods due to the lack of a multi-viewconsistency prior. To address this, we explored a more challenging task:Incomplete Multi-View Anomaly Detection (IMVAD), in which someareas of each view were masked. We proposed a pipeline for automati-cally generating the IMVAD dataset and generated the RIMAD datasetbased on the Real-IAD dataset through this pipeline. In addition, in or-der to effectively utilize the information of multiple views in the absenceof view information, we propose IMMoE, which consists of two keymodules: (1) Multi-View Expert Fusion (MVEF) effectively fuses multi-view information through a multi-view expert network and guides thereconstruction of a single view; (2) Local Anomaly Enhancement En-coder (LAEE) effectively prevents the model from overfitting the maskregion by applying dropout to local features. Our method achieves state-of-the-art performance on both the RIMAD and Real-IAD datasets, es-pecially on RIMAD, we have increased the pixel-level and image-levelmetrics by 11.8% and 2.8%, respectively. Our source code is available athttps://github.com/HULEI7/IMMoE.