Currently, deep learning-based visual inspection has been highly successful with the help of supervised learning methods. However, in real industrial scenarios, the scarcity of defect samples, the cost of annotation, and the lack of $a$ priori knowledge of defects may render supervised-based methods ineffective. In recent years, unsupervised anomaly localization (AL) algorithms have become more widely used in industrial inspection tasks. This article aims to help researchers in this field by comprehensively surveying recent achievements in unsupervised AL in industrial images using deep learning. The survey reviews more than 120 significant publications covering different aspects of AL, mainly covering various concepts, challenges, taxonomies, benchmark datasets, and quantitative performance comparisons of the methods reviewed. In reviewing the achievements to date, this article provides detailed predictions and analysis of several future research directions. This review provides detailed technical information for researchers interested in industrial AL and who wish to apply it to the localization of anomalies in other fields.

In industrial quality assessment, monitoring whether the textured product contains defects is a critical step. Compared to a large number of defect-free images that are easy to obtain, anomaly samples are limited and vary randomly in size and type. It is challenging to develop an automatic and accurate texture defect localization system that only uses normal images for training. In this paper, a multi-resolution feature learning network is proposed to detect various texture defects in an unsupervised manner. A robust pre-trained model is first employed to extract the perceptual features from the input image, then the perceptual features of various layers are fed to the corresponding multi-scale autoencoder framework. This hierarchical alignment strategy aids in receiving multi-level information for locating anomalies of various sizes. Moreover, a residual attention module (RAM) is embedded in the architecture to further improve the detection performance. Our proposed method has achieved state-of-the-art performance on the texture dataset of MVTecAD. We also extended the experiment to the real industrial texture datasets, and its detection result is better than the major existing advanced techniques.

Currently, in the field of anomaly detection, most existing methods rely on small models focus on specific industrial scenarios which exhibit strong task orientation but lack the generalization capabilities. Segment Anything Model (SAM), a vision foundation model designed for semantic segmentation tasks, has demonstrated remarkable performance in natural scene segmentation. However, SAM lacks domain-specific knowledge of industrial defect and relies on an interactive inference framework, which restricts its application in industrial anomaly detection. In this paper, we propose a novel framework: A Large-Small Model Collaboration Framework for Unsupervised Industrial Anomaly Detection (LScAD), aiming to use task-oriented small models to guide SAM for precise anomaly localization. Specifically, the small model generates the initial guidance information, which serves as the input to a multimodal prompt module. This module consists of two modalities: image and text prompt. These prompts are then used to guide SAM for accurate anomaly segmentation. Moreover, we design a dual-branch adapter to enhance SAM’s domain-specific capability through a color-domain branch and a frequency-domain branch, aiming to improve its performance in anomaly detection tasks. Extensive experiments on the MVTec AD benchmark and other real-world industrial datasets demonstrate that our method achieves state-of-the-art performance, with an image-level AUROC of 99.6%, pixel-level AUROC of 98.4%, and AP of 74.1% on the MVTec AD dataset. Our proposed method can also effortlessly adapt to multi-class anomaly detection without any modifications and achieve remarkable performance. Our code is available at: https://github.com/qsc1103/LScAD.