Industrial Anomaly Detection Based on Improved Diffusion Model: A Review

Abstract

As a class of highly effective generative models, diffusion models have attracted considerable attention in recent years and have been extensively applied to industrial anomaly detection tasks. In this review, a comprehensive discussion is presented on industrial anomaly detection based on improved diffusion models. Initially, a brief introduction to diffusion models and anomaly detection is provided, covering fundamental concepts, widely used datasets, and evaluation metrics. Recent advancements in diffusion models are then outlined from three key perspectives: inference speed, generalization ability, and reconstruction quality. Furthermore, their applications in industrial anomaly detection are examined, including sample generation, data augmentation, and the reconstruction of anomalous images. Finally, recent developments in diffusion models are summarized, and several potential research directions are suggested for future investigation.