Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Class Anomaly Detection based on Regularized Discriminative Coupled hypersphere-based Feature Adaptation

Mehdi Rafiei, Alexandros Iosifidis

TL;DR

This work tackles multi-class anomaly detection across diverse product categories by enhancing CFA with class-discriminative cues from a modified Regularized Discriminative VAE (RD-VAE), resulting in Regularized Discriminative CFA (RD-CFA). RD-CFA processes patch features through a discriminator to obtain class-aware statistics, weighs inter-class repulsion by a dissimilarity matrix, and augments the memory bank with discriminative features, all optimized via a combined loss. Extensive experiments on MVTec AD and BeanTech AD show RD-CFA outperforms eight strong baselines in both anomaly detection and localization, demonstrating effective cross-class generalization without requiring per-class models. The approach offers practical impact for scalable quality inspection in manufacturing by delivering accurate, class-aware anomaly localization with a single, unified model.

Abstract

In anomaly detection, identification of anomalies across diverse product categories is a complex task. This paper introduces a new model by including class discriminative properties obtained by a modified Regularized Discriminative Variational Auto-Encoder (RD-VAE) in the feature extraction process of Coupled-hypersphere-based Feature Adaptation (CFA). By doing so, the proposed Regularized Discriminative Coupled-hypersphere-based Feature Adaptation (RD-CFA), forms a solution for multi-class anomaly detection. By using the discriminative power of RD-VAE to capture intricate class distributions, combined with CFA's robust anomaly detection capability, the proposed method excels in discerning anomalies across various classes. Extensive evaluations on multi-class anomaly detection and localization using the MVTec AD and BeanTech AD datasets showcase the effectiveness of RD-CFA compared to eight leading contemporary methods.

Multi-Class Anomaly Detection based on Regularized Discriminative Coupled hypersphere-based Feature Adaptation

TL;DR

This work tackles multi-class anomaly detection across diverse product categories by enhancing CFA with class-discriminative cues from a modified Regularized Discriminative VAE (RD-VAE), resulting in Regularized Discriminative CFA (RD-CFA). RD-CFA processes patch features through a discriminator to obtain class-aware statistics, weighs inter-class repulsion by a dissimilarity matrix, and augments the memory bank with discriminative features, all optimized via a combined loss. Extensive experiments on MVTec AD and BeanTech AD show RD-CFA outperforms eight strong baselines in both anomaly detection and localization, demonstrating effective cross-class generalization without requiring per-class models. The approach offers practical impact for scalable quality inspection in manufacturing by delivering accurate, class-aware anomaly localization with a single, unified model.

Abstract

In anomaly detection, identification of anomalies across diverse product categories is a complex task. This paper introduces a new model by including class discriminative properties obtained by a modified Regularized Discriminative Variational Auto-Encoder (RD-VAE) in the feature extraction process of Coupled-hypersphere-based Feature Adaptation (CFA). By doing so, the proposed Regularized Discriminative Coupled-hypersphere-based Feature Adaptation (RD-CFA), forms a solution for multi-class anomaly detection. By using the discriminative power of RD-VAE to capture intricate class distributions, combined with CFA's robust anomaly detection capability, the proposed method excels in discerning anomalies across various classes. Extensive evaluations on multi-class anomaly detection and localization using the MVTec AD and BeanTech AD datasets showcase the effectiveness of RD-CFA compared to eight leading contemporary methods.
Paper Structure (11 sections, 10 equations, 6 figures, 6 tables)

This paper contains 11 sections, 10 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Coupled-hypersphere Feature Adaptation (CFA)
  4. Regularized Discriminator
  5. Regularized Discriminative CFA (RD-CFA)
  6. Evaluation
  7. Experimental Setup
  8. Quantitative Results
  9. Ablation Studies
  10. Qualitative Results
  11. Conclusions

Figures (6)

  • Figure 1: a) Single-class anomaly detection, b) Multi-class anomaly detection.
  • Figure 2: Overall structure of the CFA model.
  • Figure 3: Forces caused by reconstruction, KLD, and repulsive losses in RD-VAE.
  • Figure 4: Overall structure of the proposed RD-CFA model.
  • Figure 5: Normal, abnormal, and ground truth samples from MVTec AD and BeanTech AD datasets.
  • ...and 1 more figures