关键词: 计算机视觉, 无监督异常检测, 特征嵌入, 记忆库, 语义分割

Abstract: Unsupervised anomaly detection methods based on feature embedding often use patch-level features to localize anomalies. Although patch-level features are competitive in image-level anomaly detection tasks, there is still a lack of accuracy in pixel-level localization. To address this issue, propose MemAD, a pixel-level anomaly detection method composed of a multi-scale memory bank and a segmentation network. First, a pre-trained feature extraction network is used to extract features from normal samples in the training set, constructing a feature memory bank at three different scales. Then, during the training of the segmentation network, simulated pseudo-anomaly sample features are used to calculate the difference features with the nearest normal sample features in the memory bank, further guiding the segmentation network to learn how to locate anomalous pixels. Experimental results show that MemAD achieves image-level and pixel-level AUROC scores of 0.98 and 0.974, respectively, on the MVTec dataset, outperforming most existing methods and confirming its accuracy in pixel-level anomaly localization. This method provides a new approach and technique for unsupervised anomaly detection and has a wide range of applications.

Key words: computer vision, unsupervised anomaly detection, feature embedding, memory bank, semantic segmentation