MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing

Papers with Code Trending 09/26/25, 10:45 AM Papers

document-parsing vision-language-model open-source layout-analysis ocr computer-vision

Summary

MinerU2.5 is a 1.2B-parameter vision-language model that achieves state-of-the-art document parsing accuracy with high computational efficiency using a coarse-to-fine parsing strategy.

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.

Original Article

View Cached Full Text

Cached at: 05/08/26, 08:38 AM

Paper page - MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing

Source: https://huggingface.co/papers/2509.22186 Published on Sep 26, 2025

Submitted byhttps://huggingface.co/taesiri

taesirion Sep 29, 2025

#2 Paper of the day Authors:

Abstract

MinerU2.5, a 1.2B-parameter document parsing vision-language model, achieves state-of-the-art recognition accuracy with computational efficiency through a coarse-to-fine parsing strategy.

We introduce MinerU2.5, a 1.2B-parameterdocument parsingvision-language model that achieves state-of-the-art recognition accuracy while maintaining exceptional computational efficiency. Our approach employs acoarse-to-fine,two-stage parsingstrategy that decouples globallayout analysisfrom localcontent recognition. In the first stage, the model performs efficient layout analysis ondownsampled imagesto identify structural elements, circumventing thecomputational overheadof processing high-resolution inputs. In the second stage, guided by the global layout, it performs targetedcontent recognitiononnative-resolution cropsextracted from the original image, preserving fine-grained details in dense text, complex formulas, and tables. To support this strategy, we developed a comprehensivedata enginethat generates diverse, large-scale training corpora for bothpretrainingandfine-tuning. Ultimately, MinerU2.5 demonstrates strongdocument parsingability, achievingstate-of-the-art performanceon multiple benchmarks, surpassing both general-purpose and domain-specific models across various recognition tasks, while maintaining significantly lowercomputational overhead.