Self-supervised enhanced denoising diffusion for anomaly detection

Shu Li, Jiong Yu, Yi Lu, Guangqi Yang, Xusheng Du, Su Liu

Abstract: Highlights•Modeling normal data distribution serves as a benchmark for evaluating anomalies.•Discriminative data representation helps the model effectively detect anomalies.•Fine-grained data reconstruction provides a reliable baseline for anomaly detection.•Simplified model architecture provides enhanced model convergence.•Extracting valuable insights from limited training samples improving performance.