What is MonkeyOCR
MonkeyOCR is a document parsing model jointly launched by Huazhong University of Science and Technology and Kingsoft Office. The model supports efficiently converting unstructured document content into structured information. Based on precise layout analysis, content recognition, and logical ordering, it significantly improves the accuracy and efficiency of document parsing. Compared with traditional methods, MonkeyOCR performs excellently when processing complex documents (such as documents containing formulas and tables), with an average performance improvement of 5.1%, and improvements of 15.0% and 8.6% in formula and table parsing, respectively. The model performs excellently in multi-page document processing speed, reaching 0.84 pages per second, far exceeding other similar tools. MonkeyOCR supports multiple document types, including academic papers, textbooks, and newspapers, and is applicable to multiple languages, providing strong support for document digitization and automated processing.
Main Features of MonkeyOCR
- Document Parsing and Structuring: Converts unstructured content (including text, tables, formulas, images, etc.) in documents of various formats (such as PDF, images, etc.) into structured machine-readable information.
- Multilingual Support: Supports multiple languages, including Chinese and English.
- Efficient Processing of Complex Documents: Performs excellently when processing complex documents (such as those containing formulas, tables, multi-column layouts, etc.).
- Fast Multi-Page Document Processing: Efficiently processes multi-page documents, with a processing speed of 0.84 pages per second, significantly better than other tools (such as MinerU at 0.65 pages per second and Qwen2.5-VL-7B at 0.12 pages per second).
- Flexible Deployment and Expansion: Supports efficient deployment on a single NVIDIA 3090 GPU, meeting the needs of different scales.
Technical Principles of MonkeyOCR
- Structure-Recognition-Relation (SRR) Triple Paradigm: Based on a YOLO-based document layout detector, it identifies the positions and categories of key elements in documents (such as text blocks, tables, formulas, images, etc.). It performs content recognition on each detected region, using a large multimodal model (LMM) for end-to-end recognition to ensure high accuracy. Based on a block-level reading order prediction mechanism, it determines the logical relationships between detected elements and reconstructs the semantic structure of the document.
- MonkeyDoc Dataset: MonkeyDoc is the most comprehensive document parsing dataset to date, containing 3.9 million instances and covering more than ten document types in Chinese and English. The dataset is built based on a multi-stage pipeline, integrating careful manual annotation, programmatic synthesis, and model-driven automatic annotation. It is used to train and evaluate the MonkeyOCR model, ensuring strong generalization ability in diverse and complex document scenarios.
- Model Optimization and Deployment: Uses the AdamW optimizer and cosine learning rate scheduling, combined with large-scale datasets for training, ensuring a balance between accuracy and efficiency. Based on the LMDeploy tool, MonkeyOCR can run efficiently on a single NVIDIA 3090 GPU, supporting fast inference and large-scale deployment.
MonkeyOCR Project Address
- GitHub Repository：https://github.com/Yuliang-Liu/MonkeyOCR
- HuggingFace Model Library：https://huggingface.co/echo840/MonkeyOCR
- arXiv Technical Paper：https://arxiv.org/pdf/2506.05218
- Online Experience Demo：http://vlrlabmonkey.xyz:7685/
Application Scenarios of MonkeyOCR
- Automated Business Processes: Internal enterprise document processing, such as contracts, reports, invoices, etc., enabling automatic data extraction and structuring, improving efficiency, and reducing manual intervention.
- Digital Archiving: Institutions such as libraries and archives digitize paper documents for archiving, facilitating long-term preservation and retrieval.
- Smart Education: Educational institutions parse textbooks, exam papers, academic papers, etc., and extract content for use in online learning platforms or teaching resource repositories.
- Medical Records Management: Hospitals parse medical documents such as medical records and examination reports, and extract key information for electronic medical record systems, improving data management efficiency.
- Academic Research: Researchers parse large volumes of academic literature and extract key information for literature reviews and data analysis, assisting research work.