A Multi-scale Normalizing Flow Network for Surface Defects Detection

Surface defect detection has gained significant importance in driving intelligent manufacturing processes, improving product quality, and ensuring process stability. However, the scarcity of labeled data and the intrinsic uncertainty of anomalies continue to pose challenges in defect detection and localization. Although normalizing flow models have been developed to address these challenges, they are not suitable for industrial environments with high real-time requirements as they contain a large number of parameters that need to be trained and inferred. In order to improve the computational efficiency of normalizing flow models, we propose a new multi-scale normalizing flow network architecture for surface defect detection. The proposed multi-scale normalizing flow network can effectively reduce inference time while maintaining a certain level of detection accuracy. The effectiveness of the proposed method is validated through the real-world industrial dataset, and the application results demonstrated that the proposed method satisfies the requirement of real-time detection of surface defects in industrial environments.