pdf2md-converter 0.1.0

A PDF to clean, pagewise Markdown converter.

pdf2md: PDF to Markdown Converter

pdf2md is an open-source Python library and CLI tool for converting PDF documents into clean, page-wise Markdown files. It leverages modern, GPU-accelerated OCR and layout detection models from the Hugging Face ecosystem, with robust fallbacks to widely used tools like EasyOCR and Tesseract.

[Image of a PDF document on the left being converted to a Markdown document on the right with images and text blocks]

Features

• Multi-Backend OCR: Supports TrOCR (Hugging Face), EasyOCR, and Tesseract.
• Layout-Aware: Uses layoutparser (optional) to intelligently detect text, image, and table blocks, with a heuristic fallback.
• Resource-Aware: Explicitly manages GPU memory and resources to prevent leaks.
• CLI & Library: Use it as a powerful command-line tool or integrate it into your Python projects.
• Docker Support: CPU-only Docker image is provided by default, with clear instructions for GPU acceleration.

Quickstart (CLI)

1. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate
   

2. Install the library and dependencies:

   pip install ".[all]" # Installs all optional dependencies for full functionality
   

3. Run a conversion:

   # Convert a scanned document using pytesseract (CPU-only)
   pdf2md --input documents/scanned_book.pdf --out output/ --backend pytesseract --layout heuristic
   

Docker Usage

A Dockerfile is provided for running pdf2md in a containerized environment. By default, it's configured for CPU-only mode.

1. Build the CPU image:

   docker build -t pdf2md .
   

2. Run the CLI (CPU-only):

   docker run --rm -v $(pwd):/data pdf2md --input /data/sample.pdf --out /data/output --backend easyocr
   

How to Enable GPU Support

To run pdf2md with GPU acceleration, you need to use a base image with CUDA and install the correct PyTorch wheel.

1. Modify Dockerfile: Uncomment the FROM nvidia/cuda:11.8.0-base-ubuntu22.04 and WORKDIR /app lines, and comment out the CPU base image.
2. Update PyTorch Installation: Change the PyTorch install command to point to a CUDA-enabled wheel, for example: pip install torch==2.1.0+cu118 --index-url https://download.pytorch.org/whl/cu118. Note: Check the official PyTorch website for the latest compatible wheel URL.
3. Use docker-compose: The docker-compose.yml is pre-configured to enable NVIDIA container runtime support. Uncomment the runtime: nvidia line under the pdf2md service.
4. Build and run (GPU):

   # Build with new Dockerfile
   docker build -t pdf2md:gpu .
   # Run with docker-compose
   docker-compose up
   

Contributing & Testing

Running Tests

To run the unit and integration tests, first install the test dependencies:

pip install ".[test]"
pytest