Removing Geometric Bias in One-Class Anomaly Detection with Adaptive Feature Perturbation

2025-03-07

AI Summary

Existing one-class anomaly detection methods suffer from poor generalization due to geometric bias in benchmark datasets, while prevailing pseudo-anomaly generation techniques fail to faithfully model the intrinsic structure of normal data and over-rely on image-domain operations. Method: This paper proposes a novel paradigm that synthesizes pseudo-anomalies exclusively within a frozen pre-trained feature space—bypassing image-level augmentation entirely. Contribution/Results: Key innovations include (1) an adaptive linear feature perturbation mechanism that dynamically tailors noise distribution per sample, and (2) a contrastive learning objective explicitly decoupling geometric bias from semantic anomaly modeling. Evaluated on both standard and geometric-bias-mitigated benchmarks, our method consistently outperforms state-of-the-art approaches, demonstrating superior generalization and robustness. The implementation is publicly available.

Technology Category

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

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

One-class anomaly detection aims to detect objects that do not belong to a predefined normal class. In practice training data lack those anomalous samples; hence state-of-the-art methods are trained to discriminate between normal and synthetically-generated pseudo-anomalous data. Most methods use data augmentation techniques on normal images to simulate anomalies. However the best-performing ones implicitly leverage a geometric bias present in the benchmarking datasets. This limits their usability in more general conditions. Others are relying on basic noising schemes that may be suboptimal in capturing the underlying structure of normal data. In addition most still favour the image domain to generate pseudo-anomalies training models end-to-end from only the normal class and overlooking richer representations of the information. To overcome these limitations we consider frozen yet rich feature spaces given by pretrained models and create pseudo-anomalous features with a novel adaptive linear feature perturbation technique. It adapts the noise distribution to each sample applies decaying linear perturbations to feature vectors and further guides the classification process using a contrastive learning objective. Experimental evaluation conducted on both standard and geometric bias-free datasets demonstrates the superiority of our approach with respect to comparable baselines. The codebase is accessible via our public repository.

Problem

• Address geometric bias in one-class anomaly detection.
• Improve pseudo-anomaly generation using adaptive feature perturbation.
• Enhance anomaly detection with contrastive learning and pretrained models.

Innovation

• Adaptive linear feature perturbation technique
• Contrastive learning objective guidance
• Utilization of pretrained model feature spaces