An Unsupervised Method for Industrial Image Anomaly Detection with Vision Transformer-Based Autoencoder.

Existing industrial image anomaly detection techniques predominantly utilize codecs based on convolutional neural networks (CNNs). However, traditional convolutional autoencoders are limited to local features, struggling to assimilate global feature information. CNNs' generalizability enables the reconstruction of certain anomalous regions.

Knowledge distillation based on multi-layer fusion features.

Knowledge distillation improves the performance of a small student network by promoting it to learn the knowledge from a pre-trained high-performance but bulky teacher network. Generally, most of the current knowledge distillation methods extract relatively simple features from the middle or bottom layer of teacher network for knowledge transmission. However, the above methods ignore the fusion of features, and the fused features contain richer information.