A Masked Reverse Knowledge Distillation Method Incorporating Global and Local Information for Image Anomaly Detection

TL;DR

This paper introduces masked reverse knowledge distillation (MRKD), a novel method that enhances image anomaly detection by capturing global and local information through masking strategies, reducing overgeneralization and improving detection accuracy.

Contribution

The paper proposes MRKD, a new knowledge distillation technique using image and feature masking to improve anomaly detection by capturing comprehensive image context.

Findings

Achieves 98.9% AU-ROC at image level

Achieves 98.4% AU-ROC at pixel level

Demonstrates robustness against overgeneralization

Abstract

Knowledge distillation is an effective image anomaly detection and localization scheme. However, a major drawback of this scheme is its tendency to overly generalize, primarily due to the similarities between input and supervisory signals. In order to address this issue, this paper introduces a novel technique called masked reverse knowledge distillation (MRKD). By employing image-level masking (ILM) and feature-level masking (FLM), MRKD transforms the task of image reconstruction into image restoration. Specifically, ILM helps to capture global information by differentiating input signals from supervisory signals. On the other hand, FLM incorporates synthetic feature-level anomalies to ensure that the learned representations contain sufficient local information. With these two strategies, MRKD is endowed with stronger image context capture capacity and is less likely to be overgeneralized. Experiments on the widely-used MVTec anomaly detection dataset demonstrate that MRKD achieves impressive performance: image-level 98.9% AU-ROC, pixel-level 98.4% AU-ROC, and 95.3% AU-PRO. In addition, extensive ablation experiments have validated the superiority of MRKD in mitigating the overgeneralization problem.