FAIR: Frequency-aware Image Restoration for Industrial Visual Anomaly Detection

TL;DR

FAIR introduces a frequency-aware image restoration approach that enhances industrial anomaly detection by focusing on high-frequency components, improving normal pattern reconstruction and anomaly distinguishability.

Contribution

The paper proposes a novel self-supervised image restoration method leveraging frequency biases, achieving state-of-the-art results with a simple UNet architecture.

Findings

Achieves higher detection accuracy on multiple datasets.

Effectively balances normal reconstruction fidelity and anomaly detection.

Uses only a vanilla UNet for efficient performance.

Abstract

Image reconstruction-based anomaly detection models are widely explored in industrial visual inspection. However, existing models usually suffer from the trade-off between normal reconstruction fidelity and abnormal reconstruction distinguishability, which damages the performance. In this paper, we find that the above trade-off can be better mitigated by leveraging the distinct frequency biases between normal and abnormal reconstruction errors. To this end, we propose Frequency-aware Image Restoration (FAIR), a novel self-supervised image restoration task that restores images from their high-frequency components. It enables precise reconstruction of normal patterns while mitigating unfavorable generalization to anomalies. Using only a simple vanilla UNet, FAIR achieves state-of-the-art performance with higher efficiency on various defect detection datasets. Code: https://github.com/liutongkun/FAIR.