Transfer Learning Gaussian Anomaly Detection by Fine-tuning Representations

TL;DR

This paper introduces a novel fine-tuning method for transfer learning in anomaly detection that overcomes catastrophic forgetting by modeling normal data with a Gaussian distribution and using augmentation-based validation, achieving state-of-the-art results.

Contribution

The authors propose a new fine-tuning approach that maintains pre-trained representations for anomaly detection by modeling normal data with a Gaussian distribution and employing augmentation-based validation.

Findings

Achieves state-of-the-art anomaly detection performance on MVTec dataset.

Demonstrates robustness of the fine-tuning scheme against different augmentations.

Validates the importance of the Gaussian distribution modeling for normal data.

Abstract

Current state-of-the-art anomaly detection (AD) methods exploit the powerful representations yielded by large-scale ImageNet training. However, catastrophic forgetting prevents the successful fine-tuning of pre-trained representations on new datasets in the semi-supervised setting, and representations are therefore commonly fixed. In our work, we propose a new method to overcome catastrophic forgetting and thus successfully fine-tune pre-trained representations for AD in the transfer learning setting. Specifically, we induce a multivariate Gaussian distribution for the normal class based on the linkage between generative and discriminative modeling, and use the Mahalanobis distance of normal images to the estimated distribution as training objective. We additionally propose to use augmentations commonly employed for vicinal risk minimization in a validation scheme to detect onset of catastrophic forgetting. Extensive evaluations on the public MVTec dataset reveal that a new state of the art is achieved by our method in the AD task while simultaneously achieving anomaly segmentation performance comparable to prior state of the art. Further, ablation studies demonstrate the importance of the induced Gaussian distribution as well as the robustness of the proposed fine-tuning scheme with respect to the choice of augmentations.