Home

Awesome

<p align="center"> <img src="assets/readme/pdf-extract-kit_logo.png" width="220px" style="vertical-align:middle;"> </p> <div align="center">

English | ็ฎ€ไฝ“ไธญๆ–‡

PDF-Extract-Kit-1.0 Tutorial

[Models (๐Ÿค—Hugging Face)] | [Models(<img src="./assets/readme/modelscope_logo.png" width="20px">ModelScope)]

๐Ÿ”ฅ๐Ÿ”ฅ๐Ÿ”ฅ MinerU: Efficient Document Content Extraction Tool Based on PDF-Extract-Kit

</div> <p align="center"> ๐Ÿ‘‹ join us on <a href="https://discord.gg/Tdedn9GTXq" target="_blank">Discord</a> and <a href="https://r.vansin.top/?r=MinerU" target="_blank">WeChat</a> </p>

Overview

PDF-Extract-Kit is a powerful open-source toolkit designed to efficiently extract high-quality content from complex and diverse PDF documents. Here are its main features and advantages:

Experience PDF-Extract-Kit now and unlock the limitless potential of PDF documents!

Note: PDF-Extract-Kit is designed for high-quality document processing and functions as a model toolbox.
If you are interested in extracting high-quality document content (e.g., converting PDFs to Markdown), please use MinerU, which combines the high-quality predictions from PDF-Extract-Kit with specialized engineering optimizations for more convenient and efficient content extraction.
If you're a developer looking to create engaging applications such as document translation, document Q&A, or document assistants, you'll find it very convenient to build your own projects using PDF-Extract-Kit. In particular, we will periodically update the PDF-Extract-Kit/project directory with interesting applications, so stay tuned!

We welcome researchers and engineers from the community to contribute outstanding models and innovative applications by submitting PRs to become contributors to the PDF-Extract-Kit project.

Model Overview

Task TypeDescriptionModels
Layout DetectionLocate different elements in a document: including images, tables, text, titles, formulasDocLayout-YOLO_ft, YOLO-v10_ft, LayoutLMv3_ft
Formula DetectionLocate formulas in documents: including inline and block formulasYOLOv8_ft
Formula RecognitionRecognize formula images into LaTeX source codeUniMERNet
OCRExtract text content from images (including location and recognition)PaddleOCR
Table RecognitionRecognize table images into corresponding source code (LaTeX/HTML/Markdown)PaddleOCR+TableMaster, StructEqTable
Reading OrderSort and concatenate discrete text paragraphsComing Soon!

News and Updates

Performance Demonstration

Many current open-source SOTA models are trained and evaluated on academic datasets, achieving high-quality results only on single document types. To enable models to achieve stable and robust high-quality results on diverse documents, we constructed diverse fine-tuning datasets and fine-tuned some SOTA models to obtain practical parsing models. Below are some visual results of the models.

Layout Detection

We trained robust Layout Detection models using diverse PDF document annotations. Our fine-tuned models achieve accurate extraction results on diverse PDF documents such as papers, textbooks, research reports, and financial reports, and demonstrate high robustness to challenges like blurring and watermarks. The visualization example below shows the inference results of the fine-tuned LayoutLMv3 model.

Formula Detection

Similarly, we collected and annotated documents containing formulas in both English and Chinese, and fine-tuned advanced formula detection models. The visualization result below shows the inference results of the fine-tuned YOLO formula detection model:

Formula Recognition

UniMERNet is an algorithm designed for diverse formula recognition in real-world scenarios. By constructing large-scale training data and carefully designed results, it achieves excellent recognition performance for complex long formulas, handwritten formulas, and noisy screenshot formulas.

Table Recognition

StructEqTable is a high efficiency toolkit that can converts table images into LaTeX/HTML/MarkDown. The latest version, powered by the InternVL2-1B foundation model, improves Chinese recognition accuracy and expands multi-format output options.

For more visual and inference results of the models, please refer to the PDF-Extract-Kit tutorial documentation.

Evaluation Metrics

Coming Soon!

Usage Guide

Environment Setup

conda create -n pdf-extract-kit-1.0 python=3.10
conda activate pdf-extract-kit-1.0
pip install -r requirements.txt

Note: If your device does not support GPU, please install the CPU version dependencies using requirements-cpu.txt instead of requirements.txt.

Note๏ผš Current Doclayout-YOLO only supports installation from pypi๏ผŒif error raises during DocLayout-YOLO installation๏ผŒplease install through pip3 install doclayout-yolo==0.0.2 --extra-index-url=https://pypi.org/simple .

Model Download

Please refer to the Model Weights Download Tutorial to download the required model weights. Note: You can choose to download all the weights or select specific ones. For detailed instructions, please refer to the tutorial.

Running Demos

Layout Detection Model

python scripts/layout_detection.py --config=configs/layout_detection.yaml

Layout detection models support DocLayout-YOLO (default model), YOLO-v10, and LayoutLMv3. For YOLO-v10 and LayoutLMv3, please refer to Layout Detection Algorithm. You can view the layout detection results in the outputs/layout_detection folder.

Formula Detection Model

python scripts/formula_detection.py --config=configs/formula_detection.yaml

You can view the formula detection results in the outputs/formula_detection folder.

OCR Model

python scripts/ocr.py --config=configs/ocr.yaml

You can view the OCR results in the outputs/ocr folder.

Formula Recognition Model

python scripts/formula_recognition.py --config=configs/formula_recognition.yaml

You can view the formula recognition results in the outputs/formula_recognition folder.

Table Recognition Model

python scripts/table_parsing.py --config configs/table_parsing.yaml

You can view the table recognition results in the outputs/table_parsing folder.

Note: For more details on using the model, please refer to thePDF-Extract-Kit-1.0 Tutorial.

This project focuses on using models for high-quality content extraction from diverse documents and does not involve reconstructing extracted content into new documents, such as PDF to Markdown. For such needs, please refer to our other GitHub project: MinerU.

To-Do List

PDF-Extract-Kit aims to provide high-quality PDF content extraction capabilities. We encourage the community to propose specific and valuable needs and welcome everyone to participate in continuously improving the PDF-Extract-Kit tool to advance research and industry development.

License

This project is open-sourced under the AGPL-3.0 license.

Since this project uses YOLO code and PyMuPDF for file processing, these components require compliance with the AGPL-3.0 license. Therefore, to ensure adherence to the licensing requirements of these dependencies, this repository as a whole adopts the AGPL-3.0 license.

Acknowledgement

Citation

If you find our models / code / papers useful in your research, please consider giving โญ and citations ๐Ÿ“, thx :)

@article{wang2024mineru,
  title={MinerU: An Open-Source Solution for Precise Document Content Extraction},
  author={Wang, Bin and Xu, Chao and Zhao, Xiaomeng and Ouyang, Linke and Wu, Fan and Zhao, Zhiyuan and Xu, Rui and Liu, Kaiwen and Qu, Yuan and Shang, Fukai and others},
  journal={arXiv preprint arXiv:2409.18839},
  year={2024}
}

@misc{zhao2024doclayoutyoloenhancingdocumentlayout,
      title={DocLayout-YOLO: Enhancing Document Layout Analysis through Diverse Synthetic Data and Global-to-Local Adaptive Perception}, 
      author={Zhiyuan Zhao and Hengrui Kang and Bin Wang and Conghui He},
      year={2024},
      eprint={2410.12628},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.12628}, 
}

@misc{wang2024unimernet,
      title={UniMERNet: A Universal Network for Real-World Mathematical Expression Recognition}, 
      author={Bin Wang and Zhuangcheng Gu and Chao Xu and Bo Zhang and Botian Shi and Conghui He},
      year={2024},
      eprint={2404.15254},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

@article{he2024opendatalab,
  title={Opendatalab: Empowering general artificial intelligence with open datasets},
  author={He, Conghui and Li, Wei and Jin, Zhenjiang and Xu, Chao and Wang, Bin and Lin, Dahua},
  journal={arXiv preprint arXiv:2407.13773},
  year={2024}
}

Star History

<a> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=opendatalab/PDF-Extract-Kit&type=Date&theme=dark" /> <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=opendatalab/PDF-Extract-Kit&type=Date" /> <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=opendatalab/PDF-Extract-Kit&type=Date" /> </picture> </a>

Related Links