Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LR-IAD:Mask-Free Industrial Anomaly Detection with Logical Reasoning

Peijian Zeng, Feiyan Pang, Zhanbo Wang, Aimin Yang

TL;DR

This work tackles industrial anomaly detection under severe class imbalance and without relying on mask annotations. It introduces LR-IAD, a mask-free framework that combines Chain-of-Thought reasoning with Group Relative Policy Optimization and dual reward signals to detect and localize defects directly from raw images. The approach achieves state-of-the-art zero-shot performance on MVTec-AD and VisA, significantly surpassing mask-dependent baselines while offering interpretable, step-by-step explanations of its decisions. The results suggest substantial practical impact for scalable, cost-effective quality control in manufacturing, with future work aimed at further improving recall, reducing false positives, and extending zero-shot generalization.

Abstract

Industrial Anomaly Detection (IAD) is critical for ensuring product quality by identifying defects. Traditional methods such as feature embedding and reconstruction-based approaches require large datasets and struggle with scalability. Existing vision-language models (VLMs) and Multimodal Large Language Models (MLLMs) address some limitations but rely on mask annotations, leading to high implementation costs and false positives. Additionally, industrial datasets like MVTec-AD and VisA suffer from severe class imbalance, with defect samples constituting only 23.8% and 11.1% of total data respectively. To address these challenges, we propose a reward function that dynamically prioritizes rare defect patterns during training to handle class imbalance. We also introduce a mask-free reasoning framework using Chain of Thought (CoT) and Group Relative Policy Optimization (GRPO) mechanisms, enabling anomaly detection directly from raw images without annotated masks. This approach generates interpretable step-by-step explanations for defect localization. Our method achieves state-of-the-art performance, outperforming prior approaches by 36% in accuracy on MVTec-AD and 16% on VisA. By eliminating mask dependency and reducing costs while providing explainable outputs, this work advances industrial anomaly detection and supports scalable quality control in manufacturing. Code to reproduce the experiment is available at https://github.com/LilaKen/LR-IAD.

LR-IAD:Mask-Free Industrial Anomaly Detection with Logical Reasoning

TL;DR

This work tackles industrial anomaly detection under severe class imbalance and without relying on mask annotations. It introduces LR-IAD, a mask-free framework that combines Chain-of-Thought reasoning with Group Relative Policy Optimization and dual reward signals to detect and localize defects directly from raw images. The approach achieves state-of-the-art zero-shot performance on MVTec-AD and VisA, significantly surpassing mask-dependent baselines while offering interpretable, step-by-step explanations of its decisions. The results suggest substantial practical impact for scalable, cost-effective quality control in manufacturing, with future work aimed at further improving recall, reducing false positives, and extending zero-shot generalization.

Abstract

Industrial Anomaly Detection (IAD) is critical for ensuring product quality by identifying defects. Traditional methods such as feature embedding and reconstruction-based approaches require large datasets and struggle with scalability. Existing vision-language models (VLMs) and Multimodal Large Language Models (MLLMs) address some limitations but rely on mask annotations, leading to high implementation costs and false positives. Additionally, industrial datasets like MVTec-AD and VisA suffer from severe class imbalance, with defect samples constituting only 23.8% and 11.1% of total data respectively. To address these challenges, we propose a reward function that dynamically prioritizes rare defect patterns during training to handle class imbalance. We also introduce a mask-free reasoning framework using Chain of Thought (CoT) and Group Relative Policy Optimization (GRPO) mechanisms, enabling anomaly detection directly from raw images without annotated masks. This approach generates interpretable step-by-step explanations for defect localization. Our method achieves state-of-the-art performance, outperforming prior approaches by 36% in accuracy on MVTec-AD and 16% on VisA. By eliminating mask dependency and reducing costs while providing explainable outputs, this work advances industrial anomaly detection and supports scalable quality control in manufacturing. Code to reproduce the experiment is available at https://github.com/LilaKen/LR-IAD.
Paper Structure (21 sections, 12 equations, 3 figures, 6 tables)

This paper contains 21 sections, 12 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Anomaly Detection Methods
  4. MLLMs-Based Anomaly Detection and Generalization
  5. Method
  6. Problem Definition
  7. Reward Functions
  8. Format Reward
  9. Focal Reward
  10. GRPO Methodology
  11. Baseline Model and Prompt Template
  12. Inference Pipeline
  13. Experiment
  14. Experiment Setting
  15. Evaluation Metrics
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Overview of the task definition.
  • Figure 2: Overview of the LR-IAD framework, illustrating its components and reasoning process for anomaly detection.
  • Figure 3: Visualization examples of anomaly detection results on the MVTec-AD and VisA datasets, showcasing normal and anomalous samples.