| 中文 | English |

pdf2mind

pdf2mind is an intelligent tool powered by large language models (LLMs) that automatically converts lengthy PDF documents into well-structured mind maps with a single click. It supports output formats including XMind, FreeMind, and SVG.

Demo Show

• Watch the Demo Video
• Result Show

Environment Dependencies

This project supports both Windows and Linux environments. macOS has not been tested yet.

To set up the environment, run the following commands:

conda create --name pdf2mind python=3.12
conda activate pdf2mind
pip install -r requirements.txt -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple

Other required dependency:

• Graphviz

For linux OS system:

$ apt install graphviz

For Mac OS system:

$ brew install graphviz

For Windows OS system:

Please refer to https://graphviz.org/download/

Help

$ python pdf2mind.py -h
usage: pdf2mind.py [-h] --pdf PDF --model MODEL --language LANGUAGE (--use-doubao | --use-qwen | --use-openai) [--chunk-size CHUNK_SIZE]
                   [--overlap-size OVERLAP_SIZE] [--max-level MAX_LEVEL] [--temperature TEMPERATURE]
                   [--only-freemind | --only-xmind | --only-svg]

Command-line parser: PDF filename + Model selection

options:
  -h, --help            show this help message and exit
  --pdf PDF             PDF filename
  --model MODEL         model name
  --language LANGUAGE   Target language (e.g., 'English', 'Chinese', 'France', etc.)
  --use-doubao          Use Doubao model
  --use-qwen            Use Qwen model
  --use-openai          Use OpenAI model
  --chunk-size CHUNK_SIZE
                        chunk size of PDF (optional, default 30000)
  --overlap-size OVERLAP_SIZE
                        overlap size of PDF (optional, default 1000)
  --max-level MAX_LEVEL
                        maximum level for mind maps (optional, default: 4)
  --temperature TEMPERATURE
                        LLM temperature (optional, default: 0.7)
  --only-freemind       Only generate FreeMind (.mm) format
  --only-xmind          Only generate XMind (.md) format
  --only-svg            Only generate SVG (.svg) format

Support

Supported Models

Vendor	Required ENV Variable
OpenAI	OPENAI_API_KEY
Qwen	DASHSCOPE_API_KEY
Doubao	ARK_API_KEY

Supported Mind Map Formats

Software	Format
XMind	.md
FreeMind	.mm
SVG	.svg

Usage

Here is an example using the Doubao large language model:

$ setx ARK_API_KEY ***key***      # On Windows
$ export ARK_API_KEY=***key***    # On Linux/macOS
$ python pdf2mind.py --pdf testdata/GreenAI-2page.pdf --language Chinese --use-doubao --model doubao-1-5-lite-32k-250115

After successful execution, mind maps in all three supported formats will be generated in the source directory.

Docker

$ docker build -t hyongtao-db/pdf2mind:0.0.1 .
$ docker run  \
  hyongtao-db/pdf2mind:0.0.1 \
  -h
usage: pdf2mind.py [-h] --pdf PDF --model MODEL --language LANGUAGE
                   (--use-doubao | --use-qwen | --use-openai)
                   [--chunk-size CHUNK_SIZE] [--overlap-size OVERLAP_SIZE]
                   [--max-level MAX_LEVEL] [--temperature TEMPERATURE]
                   [--only-freemind | --only-xmind | --only-svg]

Command-line parser: PDF filename + Model selection

$ docker run \
  -e ARK_API_KEY=$ARK_API_KEY \
  -v $(pwd)/:/data/ \
  hyongtao-db/pdf2mind:0.0.1 \
  --pdf /data/testdata/GreenAI-2Page.pdf \
  --language Chinese \
  --use-doubao \
  --model doubao-1-5-lite-32k-250115

Please pay attention to the correspondence of Docker file mounts.

TODO List

• Highest Priority

  • ✅ Implement asynchronous I/O
  • ✅ Design a more comprehensive class structure
  • ✅ Add logging functionality
  • ✅ Add .gitignore
  • ✅ Optimize configuration parameters, including: model temperature, PDF chunk/overlap length, maximum depth of mind map, etc.

• Lower Priority

  • Switch to poetry for dependency management
  • ✅ Add Docker support
  • Provide a Flask-based frontend API service
  • Add unit tests using pytest
  • Add GitHub workflow automation
  • Support more models and output formats

There’s so much more I’d love to implement. Once these tasks are completed, I plan to revisit foundational knowledge on large language models and Python project best practices.

Acknowledgments

• I’ve learned a great deal from yihong0618’s xiaogpt and bilingual_book_maker.
• The project ChatPaper2Xmind has also been a great inspiration.