Flow-based one-class anomaly detection with Multi-frequency Feature fusion

Anomaly detection in computer vision seeks to identify samples outside of a predefined distribution, including texture defect detection and semantic anomaly detection. However, existing methods are limited in both types, particularly for the latter. To address this issue, we introduce the normalizing flow-based anomaly detection method. Firstly, we use semantic features extracted from a pre-trained feature extractor as input to learn the distribution of normal data from a semantic perspective. Secondly, we design a new feature fusion module based on an existing attention mechanism, which combines texture and semantic features from the feature extractor to improve the fit of the distribution function to the input data, thus enabling the model to detect both semantic and texture anomalies simultaneously. Extensive experiments on multiple well-known datasets demonstrate that our proposed method achieves state-of-the-art performance.