Outlier Exposure in Visual Industrial Anomaly Detection - Survey

Industrial anomaly detection (IAD) is crucial in safeguarding product quality and enhancing production efficiency. However, methods relying solely on positive sample training often struggle with the complexity and variability of anomalies in real-world industrial environments. Outlier Exposure (OE) approaches aim to improve model robustness by exposing the detection system to outlier samples during training, allowing the system to better differentiate between normal and abnormal conditions. This survey provides a comprehensive review of OE strategies, emphasizing their significance and potential applications in IAD. A detailed analysis is presented on how OE enhances IAD through data and feature augmentation. Moreover, advantages and challenges associated with OE in industrial contexts are discussed, along with promising future research directions. Finally, we summarize the research findings and point out future research directions.