Industrial Image Anomaly Detection using Normalizing Flow and Self-Attention

Abstract

Anomalies in industrial images can be categorized into logical anomalies and structural anomalies. Logical anomalies refer to irregularities such as object deficiency, excess, or misplacement, while structural anomalies indicate impurities such as dirt, scratches, or foreign matter inclusion. Conventional anomaly detection by normalizing flow efficiently detects structural anomalies by learning the local feature distribution of images. However, these existing methods have struggled with the detection of logical anomalies. In this study, to address this problem, we introduced a self-attention mechanism into the normalizing flow to capture global features during variable transformation. Our proposed method was evaluated and compared with a baseline normalizing flow using conventional convolution layers on the MVTecLOCO dataset. Unexpectedly, we observe that the use of the self-attention mechanism does not improve performance.