An Improved Approach for Generative Model-Based Product Image Anomaly Detection
Shota Nakada, Takumi Meguro, Qiangfu Zhao · 2022 IEEE International Conference on Consumer Electronics-Asia (ICCE-Asia) · 2022
Product anomaly detection is an essential problem in the industry. Researchers have tried to solve anomaly detection using machine learning to reduce the burden on human inspectors. In recent years, anomaly detection in product images using deep neural networks has been actively studied. In this paper, we propose an improved approach for anomaly detection using generative models. We experiment on the MVTec AD dataset, a real-world public dataset, and show that our method is applicable to real industrial products. We also confirm that performance improvements are achieved on several generative models.