Docling Pdf

This produces:

• full_document.md - Complete markdown with cleaned references
• figures/ - Numbered PNG files (figure_001.png, figure_002.png, etc.)
• tables/ - Individual markdown tables (table_001.md, table_002.md, etc.)
• metadata.json - Document statistics and conversion timing

Conversion Process

Step 1: Prepare the Environment

The script uses uv to manage dependencies automatically. No manual installation required. The script's inline metadata specifies all required packages.

Step 2: Execute Conversion

Run the conversion script with the following syntax:

uv run --python 3.10 scripts/convert_pdf.py <pdf_file> <output_folder> [options]

Required arguments:

• pdf_file - Path to the input PDF file
• output_folder - Directory where output will be saved

Optional arguments:

• --image-resolution-scale F - Scale factor for extracted images (default: 2.0)

Examples:

# Basic conversion
uv run --python 3.10 scripts/convert_pdf.py paper.pdf output/

Step 3: Process Output

The conversion creates a structured output directory:

output_folder/
├── full_document.md     # Complete markdown (cleaned references)
├── figures/             # PNG images
│   ├── figure_001.png
│   ├── figure_002.png
│   └── ...
├── tables/              # Markdown tables
│   ├── table_001.md
│   ├── table_002.md
│   └── ...
└── metadata.json        # Conversion statistics

Key features:

• Figure references in markdown are automatically updated to point to figures/ directory
• Tables are exported as DataFrames converted to markdown format
• Original artifacts folders are cleaned up automatically
• Metadata includes page count, timing, figure/table counts

Script Details

The scripts/convert_pdf.py script is a standalone Python script with inline dependencies that:

Important: The script is designed to be run with uv run which handles environment creation and dependency management automatically. Do not try to run it directly with python3 without first installing dependencies.