# MAU-GPT: Enhancing Multi-type Industrial Anomaly Understanding via Anomaly-aware and Generalist Experts Adaptation

Source: [MAU-GPT: Enhancing Multi-type Industrial Anomaly Understanding via Anomaly-aware and Generalist Experts Adaptation](<https://www.sota2.com/research/paper/698abba8c97a0b3e3b5ddfe6>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-14T05:00:04.348Z

Authors: Zhuonan Wang, Zhenxuan Fan, Siwen Tan, Yu Zhong, Yuqian Yuan, Haoyuan Li, Hao Jiang, Wenqiao Zhang, Feifei Shao, Hongwei Wang, Jun Xiao

Published: 2026-01-31T05:36:49.000Z

## Abstract

As industrial manufacturing scales, automating fine-grained product image analysis has become critical for quality control. However, existing approaches are hindered by limited dataset coverage and poor model generalization across diverse and complex anomaly patterns. To address these challenges, we introduce MAU-Set, a comprehensive dataset for Multi-type industrial Anomaly Understanding. It spans multiple industrial domains and features a hierarchical task structure, ranging from binary classification to complex reasoning. Alongside this dataset, we establish a rigorous evaluation protocol to facilitate fair and comprehensive model assessment. Building upon this foundation, we further present MAU-GPT, a domain-adapted multimodal large model specifically designed for industrial anomaly understanding. It incorporates a novel AMoE-LoRA mechanism that unifies anomaly-aware and generalist experts adaptation, enhancing both understanding and reasoning across diverse defect classes. Extensive experiments show that MAU-GPT consistently outperforms prior state-of-the-art methods across all domains, demonstrating strong potential for scalable and automated industrial inspection.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Discriminative QA on MMAD](<https://www.sota2.com/research/sota/discriminative-qa-on-mmad.md>): Discriminative QA; MMAD; Industrial Components Score: 89.37.

- [Open-Ended QA on MMAD](<https://www.sota2.com/research/sota/open-ended-qa-on-mmad.md>): Open-ended QA; MMAD; Industrial Components Score: 88.75.

- [Open-Ended Question Answering on MMAD (test)](<https://www.sota2.com/research/sota/open-ended-question-answering-on-mmad-test.md>): Open-ended Question Answering; MMAD (test); ROUGE-1: 0.7026.

- [Industrial Anomaly Detection on MMAD](<https://www.sota2.com/research/sota/industrial-anomaly-detection-on-mmad.md>): Industrial Anomaly Detection; MMAD; Anomaly Discrimination: 67.5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2602.07011>)

[PDF](<https://arxiv.org/pdf/2602.07011.pdf>)