MOCR - A Multimodal Document Parsing Model Jointly Launched by Xiaohongshu and Huazhong University of Science and Technology

What is MOCR

MOCR (Multimodal OCR) is a multimodal document parsing model jointly launched by Huazhong University of Science and Technology and Xiaohongshu hi lab. With only 3B parameters, it achieves breakthrough performance in document parsing and graphics reconstruction. The model breaks the limitation of traditional OCR that only recognizes text, parsing visual elements such as charts, formulas, and flowcharts into editable SVG code, realizing a new paradigm of "parsing everything." It ranks first among open-source models, and its graphics reconstruction capability even surpasses Gemini 3 Pro, bringing a paradigm shift to the field of document AI.

MOCR's Main Features

• Full-element document parsing: Supports recognition of all page elements including text, tables, formulas, and charts, outputs structured data and maintains reading order.
• Graphics to SVG code: Reconstructs visual content such as statistical charts, scientific illustrations, and UI layouts into editable SVG code.
• Multi-format input support: Supports parsing of multiple document types such as PDFs, webpage screenshots, scanned documents, and mobile phone photos.
• General visual capabilities: The model has general multimodal understanding capabilities such as visual question answering, visual grounding, and image captioning.
• Dual-version models: Provides a balanced version dots.mocr and an SVG-optimized version dots.mocr-svg to meet the needs of different scenarios.

MOCR's Key Information and Usage Requirements

• Development team: Huazhong University of Science and Technology × Xiaohongshu hi lab
• Model parameters: 3B (1.2B vision encoder + 1.5B language decoder)
• Model versions: dots.mocr (balanced version), dots.mocr-svg (SVG-enhanced version)
• Core innovation: Parses graphics into SVG code, realizing a new paradigm of "parsing everything"
• Performance: First in open-source document parsing, graphics reconstruction surpasses Gemini 3 Pro
• GPU: CUDA-supported NVIDIA graphics card (recommended for inference acceleration)
• Memory: Adjusted according to input resolution; high-resolution documents require larger VRAM

MOCR's Core Advantages

• Small parameters, big capabilities: With only 3B parameters, its performance surpasses many large models, ranking first in open-source document parsing and surpassing Gemini 3 Pro in graphics reconstruction.
• Full-element parsing: The model breaks the limitation of traditional OCR that only recognizes text, uniformly parsing visual elements such as charts, formulas, and flowcharts into structured code.
• Editable graphics: Converts graphics into SVG code, enabling lossless reconstruction and secondary editing, rather than simply cropping into pixel images.
• Data engine innovation: Supports building multi-source data pipelines such as PDFs, webpages, and SVG assets, solving the problem of scarce graphics supervision signals.
• Evaluation method innovation: The model proposes the OCR Arena framework, using strong VLMs as judges for reliable comparative evaluation.

How to Use MOCR

• Environment preparation: Create a Python 3.12 virtual environment, clone the GitHub repository, and install dependencies.
• Download the model: Run the download script to obtain the model weights; note that the save path should not contain English periods.
• Start the service: Deploy the model service using vLLM, supporting GPU-accelerated inference.
• Document parsing: Call the parsing script to process images or PDF files and output structured results.
• Graphics Conversion: Use a dedicated SVG script to convert charts into editable SVG code.
• Get Results: Obtain the generated JSON file containing bounding boxes, Markdown text, and visualized annotation images.

MOCR Project Address

• GitHub Repository：https://github.com/rednote-hilab/dots.mocr
• arXiv Technical Paper：https://arxiv.org/pdf/2603.13032
• Online Demo：https://dotsocr.xiaohongshu.com/

Comparison of MOCR with Similar Competitors

MOCR Application Scenarios

• Academic Research: Parse paper PDFs, extract formulas, reconstruct charts, convert scanned papers into editable LaTeX, and reproduce research chart data.
• Finance and Economics: Analyze financial reports, extract data tables, digitize charts, and convert bar charts and line charts in PDF financial reports into data usable in Excel.
• Law and Government Affairs: Review contracts, digitize case files, recognize certificates, and structurally extract key information from multi-page contracts while maintaining complete formatting.
• Education and Publishing: Digitize textbooks, build question banks, recognize blackboard writing, and convert complex formulas and charts in printed textbooks into electronic resources.
• Healthcare: Parse medical records, extract lab reports, annotate medical images, and process documents containing chemical molecular formulas and medical diagrams.