Memory-Distilled Selection for Noise-Robust Anomaly Detection

2026-05-26

AI Summary

This work addresses the performance degradation in unsupervised industrial anomaly detection caused by contaminated training data. To this end, the authors propose Memory-Distilled Selection (MeDS), a method that constructs a sparsely ensembled memory bank via random subsampling to coarsely identify corrupted samples. The aggregated distance metrics from this memory bank are then distilled into a reconstruction scoring network, which is subsequently fine-tuned on the selected clean data to enable fine-grained anomaly localization. Notably, MeDS requires no hyperparameter tuning for varying noise ratios and demonstrates robustness across a wide range of contamination levels, achieving stable detection even under high noise conditions. It attains a state-of-the-art image-level AUROC of 99.16% on MVTec AD with 40% label noise and sets new benchmarks on VisA and Real-IAD under noisy scenarios.

Abstract

Anomaly detection (AD) under data contamination is critical for deploying unsupervised defect detection in industrial environments, where curating perfectly clean training sets is impractical. However, existing methods are sensitive to contamination, suffering significant performance degradation as the noise ratio increases. In this paper, we propose Memory-Distilled Selection (MeDS), a training algorithm based on data selection. MeDS constructs an ensemble of partial memories via random subsampling, where the resulting sparsity acts as a low-pass filter that captures nominal patterns across a wide range of noise ratios, enabling coarse-level identification of contaminated samples. The aggregated distances to the bootstrapped memories are then distilled into a reconstruction score network, which is subsequently fine-tuned on clean data filtered using scores from the distilled model, enabling fine-grained localization of anomalies. MeDS is robust across a wide range of noise ratios without requiring noise-ratio-specific hyperparameter tuning, achieving 99.16\% image-level AUROC on MVTecAD at a 40\% noise ratio, and attaining state-of-the-art performance on both VisA and Real-IAD under noisy settings. We thoroughly verify the efficacy of MeDS on industrial AD benchmarks under noisy data scenarios, accompanied by in-depth empirical analyses.

Problem

• anomaly detection
• data contamination
• noise-robust
• unsupervised defect detection
• industrial environments

Innovation

• Memory-Distilled Selection
• noise-robust anomaly detection
• data selection
• partial memory ensemble
• reconstruction score distillation