References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: CAINNFlow: Convolutional block Attention modules and Invertible Neural Networks Flow for anomaly detection and localization tasks
(Submitted on 4 Jun 2022 (this version), latest version 15 Dec 2022 (v7))
Abstract: Detection of object anomalies is crucial in industrial processes, but unsupervised anomaly detection and localization is particularly important due to the difficulty of obtaining a large number of defective samples and the unpredictable types of anomalies in real life. Among the existing unsupervised anomaly detection and localization methods, the NF-based scheme has achieved better results. However, the two subnets (complex functions) si(ui) and ti(ui) in NF are usually multilayer perceptrons, which need to squeeze the input visual features from 2D flattening to 1D, destroying the spatial location relationship in the feature map and losing the spatial structure information. In order to retain and effectively extract spatial structure information, we design in this study a complex function model with alternating CBAM embedded in a stacked 3*3 full convolution, which is able to retain and effectively extract spatial structure information in the normalized flow model. Extensive experimental results on the MVTec AD dataset show that CAINNFlow achieves advanced levels of accuracy and inference efficiency based on CNN and Transformer backbone networks as feature extractors, and CAINNFlow achieves a pixel-level AUC of 98.76\% for anomaly detection in MVTec AD.
Submission history
From: Ruiqing Yan [view email][v1] Sat, 4 Jun 2022 13:45:08 GMT (8406kb,D)
[v2] Tue, 7 Jun 2022 17:27:44 GMT (8406kb,D)
[v3] Wed, 8 Jun 2022 16:04:37 GMT (8434kb,D)
[v4] Wed, 15 Jun 2022 17:33:30 GMT (8434kb,D)
[v5] Fri, 21 Oct 2022 14:46:35 GMT (8440kb,D)
[v6] Thu, 8 Dec 2022 05:55:29 GMT (8460kb,D)
[v7] Thu, 15 Dec 2022 10:15:34 GMT (8460kb,D)
Link back to: arXiv, form interface, contact.