Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ROADS: Robust Prompt-driven Multi-Class Anomaly Detection under Domain Shift

Hossein Kashiani, Niloufar Alipour Talemi, Fatemeh Afghah

TL;DR

ROADS employs a hierarchical class-aware prompt integration mechanism that dynamically encodes class-specific information into the authors' anomaly detector to mitigate interference among anomaly classes and incorporates a domain adapter to enhance robustness against domain shifts by learning domain-invariant representations.

Abstract

Recent advancements in anomaly detection have shifted focus towards Multi-class Unified Anomaly Detection (MUAD), offering more scalable and practical alternatives compared to traditional one-class-one-model approaches. However, existing MUAD methods often suffer from inter-class interference and are highly susceptible to domain shifts, leading to substantial performance degradation in real-world applications. In this paper, we propose a novel robust prompt-driven MUAD framework, called ROADS, to address these challenges. ROADS employs a hierarchical class-aware prompt integration mechanism that dynamically encodes class-specific information into our anomaly detector to mitigate interference among anomaly classes. Additionally, ROADS incorporates a domain adapter to enhance robustness against domain shifts by learning domain-invariant representations. Extensive experiments on MVTec-AD and VISA datasets demonstrate that ROADS surpasses state-of-the-art methods in both anomaly detection and localization, with notable improvements in out-of-distribution settings.

ROADS: Robust Prompt-driven Multi-Class Anomaly Detection under Domain Shift

TL;DR

ROADS employs a hierarchical class-aware prompt integration mechanism that dynamically encodes class-specific information into the authors' anomaly detector to mitigate interference among anomaly classes and incorporates a domain adapter to enhance robustness against domain shifts by learning domain-invariant representations.

Abstract

Recent advancements in anomaly detection have shifted focus towards Multi-class Unified Anomaly Detection (MUAD), offering more scalable and practical alternatives compared to traditional one-class-one-model approaches. However, existing MUAD methods often suffer from inter-class interference and are highly susceptible to domain shifts, leading to substantial performance degradation in real-world applications. In this paper, we propose a novel robust prompt-driven MUAD framework, called ROADS, to address these challenges. ROADS employs a hierarchical class-aware prompt integration mechanism that dynamically encodes class-specific information into our anomaly detector to mitigate interference among anomaly classes. Additionally, ROADS incorporates a domain adapter to enhance robustness against domain shifts by learning domain-invariant representations. Extensive experiments on MVTec-AD and VISA datasets demonstrate that ROADS surpasses state-of-the-art methods in both anomaly detection and localization, with notable improvements in out-of-distribution settings.

Paper Structure

This paper contains 14 sections, 6 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Problem Definition
  5. Reverse Distillation Framework
  6. Hierarchical Class-Aware Prompt Integration
  7. Domain Adapter
  8. Training Objective
  9. Experiments
  10. Experimental settings
  11. In-distribution Evaluation
  12. Out-of-Distribution Evaluation
  13. Ablation Study
  14. Conclusion

Figures (3)

  • Figure 1: Comparison of multi-class anomaly detection methods on the MVTec-AD bergmann2019mvtec in both in-distribution and out-of-distribution settings. Image-level AUROC and pixel-level AUPRO are used for anomaly detection and localization evaluations. Methods, including UniAD Unified2024, GNL cao2023anomaly, and ViTAD vitad , are evaluated under a unified setup. Our method exhibits exceptional robustness in out-of-distribution scenarios while outperforming competing approaches in in-distribution settings.
  • Figure 2: Overview of the proposed ROADS framework. First, the domain adapter aligns the features of the OOD target domain. Then, the corresponding learned class-specific prompt tokens are selected from the pool and integrated into the student decoder.
  • Figure 3: Qualitative comparison between the proposed ROADS method and UniAD Unified2024 on the MVTec-AD dataset bergmann2019mvtec under both ID and OOD settings.