DifferAD-R1: A Difference-Guided IndustrialAnomaly Localization with Multimodal LargeLanguage Models

2026-06-15

AI Summary

This work addresses the challenge of precisely localizing unseen defect categories in industrial inspection scenarios, where existing methods suffer from poor cross-scenario generalization and a paradigm mismatch when applying multimodal large language models (MLLMs) to localization tasks, compounded by insufficient optimization signals. To overcome these limitations, we propose DifferAD-R1, a novel framework that introduces a difference-guided dual-image localization paradigm, reframing anomaly detection as a single-step difference localization problem. Our approach integrates an MLLM with an enhanced GRPO reinforcement learning algorithm and incorporates a dual-consistency localization reward mechanism alongside a difficulty-aware adaptive reweighting and grouped resampling strategy, significantly enhancing sensitivity to subtle defects. Evaluated on our newly curated AD-DualDiff dataset—comprising 13K images across 20 defect categories—DifferAD-R1 substantially outperforms current methods and achieves performance comparable to large-scale models such as Qwen3-VL (235B parameters).

Abstract

Industrial anomaly localization aims to accurately identify and localize abnormal regions in industrial products, addressing the critical challenge of detecting unseen defect categories in real-world scenarios. Traditional closed-set methods often suffer from poor cross-scenario generalization, while existingMultimodal Large Language Model (MLLM)-based approachesface two core limitations: they either adopt QA-style paradigmsmisaligned with the practical demands of localization, or relyon standard optimization techniques such as Group RelativePolicy Optimization (GRPO), which fails to deliver effectivelearning signals for subtle defects. To tackle these issues, thispaper proposes DifferAD-R1, an MLLM-augmented reinforcement learning framework tailored for industrial anomaly localization. We design a Difference-Guided dual-image paradigm,which reformulates the localization task as a one-shot difference grounding problem to effectively explore cross-scenarioanomalies. A Dual-Consistency Localization Reward is developedfor hard-to-detect anomalies, enhancing optimization stabilityand robustness. Additionally, we integrate a difficulty-awarestrategy with adaptive reweighting and group-wise resamplingto prioritize learning on challenging instances. To facilitateevaluations in real-world industrial settings, we construct theAD-DualDiff dataset, comprising 13K paired images across 20categories. Experimental results demonstrate that DifferADR1 significantly outperforms existing baselines and achievescompetitive performance compared to large-scale models likeQwen3-VL (235B parameters). Our code is publicly availableat: https://github.com/Rong2026/work-1.

Problem

• industrial anomaly localization
• unseen defect categories
• cross-scenario generalization
• subtle defects
• multimodal large language models

Innovation

• Difference-Guided Paradigm
• Multimodal Large Language Model
• Anomaly Localization
• Reinforcement Learning
• Dual-Consistency Reward