Deep Industrial Image Anomaly Detection: A Survey

TL;DR

Industrial image anomaly detection aims to automatically identify defects on production lines using deep learning. This survey organizes methods by network architecture (e.g., teacher-student, memory bank, autoencoder, GANs, transformers, diffusion) and by supervision level (unsupervised vs. supervised), and it discusses four industrial settings: few-shot, noisy, 3D, and anomaly synthesis. It compiles datasets and evaluation metrics, reviews performance on standard benchmarks like MVTec AD, and analyzes practical challenges such as data realism, cross-domain deployment, and inference speed. The authors outline future directions, including multi-modal data fusion, lightweight models, and hybrid supervised–unsupervised strategies to bridge research with industrial practice. Overall, the paper provides a comprehensive taxonomy, performance insights, and a roadmap for advancing robust, scalable IAD in real manufacturing environments.

Abstract

The recent rapid development of deep learning has laid a milestone in industrial Image Anomaly Detection (IAD). In this paper, we provide a comprehensive review of deep learning-based image anomaly detection techniques, from the perspectives of neural network architectures, levels of supervision, loss functions, metrics and datasets. In addition, we extract the new setting from industrial manufacturing and review the current IAD approaches under our proposed our new setting. Moreover, we highlight several opening challenges for image anomaly detection. The merits and downsides of representative network architectures under varying supervision are discussed. Finally, we summarize the research findings and point out future research directions. More resources are available at https://github.com/M-3LAB/awesome-industrial-anomaly-detection.

Figures (14)

• Figure 1: Framework of this survey.
• Figure 2: Architecture of teacher-student models.
• Figure 3: Architecture of one-class classification models.
• Figure 4: Architecture of distribution-map based methods.
• Figure 5: Architecture of memory bank based methods.