_version_ 1866912614699761664
author Niu, Junbo
Liu, Zheng
Gu, Zhuangcheng
Wang, Bin
Ouyang, Linke
Zhao, Zhiyuan
Chu, Tao
He, Tianyao
Wu, Fan
Zhang, Qintong
Jin, Zhenjiang
Liang, Guang
Zhang, Rui
Zhang, Wenzheng
Qu, Yuan
Ren, Zhifei
Sun, Yuefeng
Zheng, Yuanhong
Ma, Dongsheng
Tang, Zirui
Niu, Boyu
Miao, Ziyang
Dong, Hejun
Qian, Siyi
Zhang, Junyuan
Chen, Jingzhou
Wang, Fangdong
Zhao, Xiaomeng
Wei, Liqun
Li, Wei
Wang, Shasha
Xu, Ruiliang
Cao, Yuanyuan
Chen, Lu
Wu, Qianqian
Gu, Huaiyu
Lu, Lindong
Wang, Keming
Lin, Dechen
Shen, Guanlin
Zhou, Xuanhe
Zhang, Linfeng
Zang, Yuhang
Dong, Xiaoyi
Wang, Jiaqi
Zhang, Bo
Bai, Lei
Chu, Pei
Li, Weijia
Wu, Jiang
Wu, Lijun
Li, Zhenxiang
Wang, Guangyu
Tu, Zhongying
Xu, Chao
Chen, Kai
Qiao, Yu
Zhou, Bowen
Lin, Dahua
Zhang, Wentao
He, Conghui
author_facet Niu, Junbo
Liu, Zheng
Gu, Zhuangcheng
Wang, Bin
Ouyang, Linke
Zhao, Zhiyuan
Chu, Tao
He, Tianyao
Wu, Fan
Zhang, Qintong
Jin, Zhenjiang
Liang, Guang
Zhang, Rui
Zhang, Wenzheng
Qu, Yuan
Ren, Zhifei
Sun, Yuefeng
Zheng, Yuanhong
Ma, Dongsheng
Tang, Zirui
Niu, Boyu
Miao, Ziyang
Dong, Hejun
Qian, Siyi
Zhang, Junyuan
Chen, Jingzhou
Wang, Fangdong
Zhao, Xiaomeng
Wei, Liqun
Li, Wei
Wang, Shasha
Xu, Ruiliang
Cao, Yuanyuan
Chen, Lu
Wu, Qianqian
Gu, Huaiyu
Lu, Lindong
Wang, Keming
Lin, Dechen
Shen, Guanlin
Zhou, Xuanhe
Zhang, Linfeng
Zang, Yuhang
Dong, Xiaoyi
Wang, Jiaqi
Zhang, Bo
Bai, Lei
Chu, Pei
Li, Weijia
Wu, Jiang
Wu, Lijun
Li, Zhenxiang
Wang, Guangyu
Tu, Zhongying
Xu, Chao
Chen, Kai
Qiao, Yu
Zhou, Bowen
Lin, Dahua
Zhang, Wentao
He, Conghui
contents We introduce MinerU2.5, a 1.2B-parameter document parsing vision-language model that achieves state-of-the-art recognition accuracy while maintaining exceptional computational efficiency. Our approach employs a coarse-to-fine, two-stage parsing strategy that decouples global layout analysis from local content recognition. In the first stage, the model performs efficient layout analysis on downsampled images to identify structural elements, circumventing the computational overhead of processing high-resolution inputs. In the second stage, guided by the global layout, it performs targeted content recognition on native-resolution crops extracted from the original image, preserving fine-grained details in dense text, complex formulas, and tables. To support this strategy, we developed a comprehensive data engine that generates diverse, large-scale training corpora for both pretraining and fine-tuning. Ultimately, MinerU2.5 demonstrates strong document parsing ability, achieving state-of-the-art performance on multiple benchmarks, surpassing both general-purpose and domain-specific models across various recognition tasks, while maintaining significantly lower computational overhead.
format Preprint
id arxiv_https___arxiv_org_abs_2509_22186
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing
Niu, Junbo
Liu, Zheng
Gu, Zhuangcheng
Wang, Bin
Ouyang, Linke
Zhao, Zhiyuan
Chu, Tao
He, Tianyao
Wu, Fan
Zhang, Qintong
Jin, Zhenjiang
Liang, Guang
Zhang, Rui
Zhang, Wenzheng
Qu, Yuan
Ren, Zhifei
Sun, Yuefeng
Zheng, Yuanhong
Ma, Dongsheng
Tang, Zirui
Niu, Boyu
Miao, Ziyang
Dong, Hejun
Qian, Siyi
Zhang, Junyuan
Chen, Jingzhou
Wang, Fangdong
Zhao, Xiaomeng
Wei, Liqun
Li, Wei
Wang, Shasha
Xu, Ruiliang
Cao, Yuanyuan
Chen, Lu
Wu, Qianqian
Gu, Huaiyu
Lu, Lindong
Wang, Keming
Lin, Dechen
Shen, Guanlin
Zhou, Xuanhe
Zhang, Linfeng
Zang, Yuhang
Dong, Xiaoyi
Wang, Jiaqi
Zhang, Bo
Bai, Lei
Chu, Pei
Li, Weijia
Wu, Jiang
Wu, Lijun
Li, Zhenxiang
Wang, Guangyu
Tu, Zhongying
Xu, Chao
Chen, Kai
Qiao, Yu
Zhou, Bowen
Lin, Dahua
Zhang, Wentao
He, Conghui
Computer Vision and Pattern Recognition
Computation and Language
We introduce MinerU2.5, a 1.2B-parameter document parsing vision-language model that achieves state-of-the-art recognition accuracy while maintaining exceptional computational efficiency. Our approach employs a coarse-to-fine, two-stage parsing strategy that decouples global layout analysis from local content recognition. In the first stage, the model performs efficient layout analysis on downsampled images to identify structural elements, circumventing the computational overhead of processing high-resolution inputs. In the second stage, guided by the global layout, it performs targeted content recognition on native-resolution crops extracted from the original image, preserving fine-grained details in dense text, complex formulas, and tables. To support this strategy, we developed a comprehensive data engine that generates diverse, large-scale training corpora for both pretraining and fine-tuning. Ultimately, MinerU2.5 demonstrates strong document parsing ability, achieving state-of-the-art performance on multiple benchmarks, surpassing both general-purpose and domain-specific models across various recognition tasks, while maintaining significantly lower computational overhead.
title MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2509.22186