Levarging Learning Bias for Noisy Anomaly Detection

2025-08-10

AI Summary

In fully unsupervised image anomaly detection (FUIAD), the absence of ground-truth labels often leads to implicit contamination of training sets with unlabeled anomalies, causing models to erroneously learn anomalies as normal. Method: This paper proposes a two-stage self-purification framework that systematically exploits inherent learning biases of deep models—namely, statistical dominance and feature-space distribution disparity—to identify and filter anomalous samples without human annotation. The method comprises multi-submodel collaborative training, cross-model anomaly score aggregation, and feature-space consistency analysis, and is compatible with mainstream unsupervised backbone architectures. Contribution/Results: Evaluated on the Real-IAD benchmark, our approach achieves significant improvements over existing state-of-the-art methods across diverse contamination levels, demonstrating strong robustness and high localization accuracy.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Adversarial Attacks & Robustness
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

This paper addresses the challenge of fully unsupervised image anomaly detection (FUIAD), where training data may contain unlabeled anomalies. Conventional methods assume anomaly-free training data, but real-world contamination leads models to absorb anomalies as normal, degrading detection performance. To mitigate this, we propose a two-stage framework that systematically exploits inherent learning bias in models. The learning bias stems from: (1) the statistical dominance of normal samples, driving models to prioritize learning stable normal patterns over sparse anomalies, and (2) feature-space divergence, where normal data exhibit high intra-class consistency while anomalies display high diversity, leading to unstable model responses. Leveraging the learning bias, stage 1 partitions the training set into subsets, trains sub-models, and aggregates cross-model anomaly scores to filter a purified dataset. Stage 2 trains the final detector on this dataset. Experiments on the Real-IAD benchmark demonstrate superior anomaly detection and localization performance under different noise conditions. Ablation studies further validate the framework's contamination resilience, emphasizing the critical role of learning bias exploitation. The model-agnostic design ensures compatibility with diverse unsupervised backbones, offering a practical solution for real-world scenarios with imperfect training data. Code is available at https://github.com/hustzhangyuxin/LLBNAD.

Problem

• Detect image anomalies in noisy unsupervised training data
• Mitigate model bias from contaminated normal samples
• Improve anomaly detection with a two-stage purification framework

Innovation

• Two-stage framework leveraging learning bias
• Cross-model anomaly scores filter dataset
• Model-agnostic design for diverse backbones