Mean-Shifted Contrastive Loss for Anomaly Detection
This work addresses anomaly detection in computer vision, offering a novel method to enhance performance by modifying contrastive learning for pre-trained features, though it is incremental as it builds on existing transfer learning approaches.
The paper tackled the problem of improving anomaly detection by fine-tuning pre-trained representations on normal images, but found that standard contrastive learning fails due to poor optimization dynamics from pre-trained feature initialization. It introduced the Mean-Shifted Contrastive Loss, achieving state-of-the-art performance with 98.6% ROC-AUC on CIFAR-10.
Deep anomaly detection methods learn representations that separate between normal and anomalous images. Although self-supervised representation learning is commonly used, small dataset sizes limit its effectiveness. It was previously shown that utilizing external, generic datasets (e.g. ImageNet classification) can significantly improve anomaly detection performance. One approach is outlier exposure, which fails when the external datasets do not resemble the anomalies. We take the approach of transferring representations pre-trained on external datasets for anomaly detection. Anomaly detection performance can be significantly improved by fine-tuning the pre-trained representations on the normal training images. In this paper, we first demonstrate and analyze that contrastive learning, the most popular self-supervised learning paradigm cannot be naively applied to pre-trained features. The reason is that pre-trained feature initialization causes poor conditioning for standard contrastive objectives, resulting in bad optimization dynamics. Based on our analysis, we provide a modified contrastive objective, the Mean-Shifted Contrastive Loss. Our method is highly effective and achieves a new state-of-the-art anomaly detection performance including $98.6\%$ ROC-AUC on the CIFAR-10 dataset.