paper2epub 0.2.0

A powerful academic PDF to EPUB converter with AI-powered layout detection and LaTeX math support

paper2epub

A powerful academic PDF to EPUB converter with AI-powered layout detection and LaTeX math support.

Features

• Academic-First Design: Optimized for scientific papers, research documents, and technical publications
• LaTeX Math Support: Preserves mathematical equations using Nougat's neural OCR
• Complex Layout Handling: AI-powered detection of multi-column layouts, tables, and figures
• GPU Acceleration: Optional CUDA/MPS (Apple Silicon) support for faster processing
• Figure Extraction: Automatic extraction and embedding of figures using PyMuPDF
• Multiple Output Formats: EPUB3 with optional intermediate Markdown
• Easy to Use: Both CLI and Python API available

Installation

Basic Installation

pip install paper2epub

From Source

git clone https://github.com/MAXNORM8650/paper2epub.git
cd paper2epub
pip install -e .

Development Installation

pip install -e ".[dev]"

Requirements

• Python 3.9+
• PyTorch 2.0+
• For GPU acceleration:
  • NVIDIA GPU: CUDA-enabled PyTorch
  • Apple Silicon (M1/M2/M3): MPS-enabled PyTorch (included by default)

Quick Start

Command Line

# Basic conversion
paper2epub paper.pdf

# Specify output and metadata
paper2epub paper.pdf -o output.epub -t "My Paper" -a "John Doe"

# Use larger model with GPU
paper2epub paper.pdf -m base -d cuda

# Save intermediate markdown
paper2epub paper.pdf --save-markdown

# Skip figure extraction
paper2epub paper.pdf --no-figures

# Set minimum figure size (filter small images)
paper2epub paper.pdf --figure-min-size 150

Python API

from paper2epub import Paper2EpubConverter

# Initialize converter
converter = Paper2EpubConverter(
    model_tag="0.1.0-small",  # or "0.1.0-base" for better quality
    device="auto",             # auto-detect GPU/CPU
    extract_figures=True,      # enable figure extraction
    figure_min_size=100,       # minimum figure size in pixels
)

# Convert PDF to EPUB
output_path = converter.convert(
    pdf_path="paper.pdf",
    title="My Academic Paper",
    author="John Doe",
    save_markdown=True,        # optionally save .md file
)

print(f"Created: {output_path}")

CLI Options

Usage: paper2epub [OPTIONS] PDF_PATH

Options:
  -o, --output PATH          Output EPUB file path
  -t, --title TEXT           Book title
  -a, --author TEXT          Author name
  -l, --language TEXT        Language code (default: en)
  -m, --model [small|base]   Nougat model size (default: small)
  -d, --device [auto|cuda|mps|cpu]  Device to use
  -b, --batch-size INT       Batch size for processing
  --save-markdown            Save intermediate markdown file
  --no-figures               Skip figure extraction from PDF
  --figure-min-size INT      Minimum figure size in pixels (default: 100)
  -v, --verbose              Enable verbose logging
  --version                  Show version
  --help                     Show this message and exit

How It Works

paper2epub uses a multi-stage pipeline:

1. PDF Extraction: Nougat (Meta's neural OCR) extracts text, tables, and LaTeX equations
2. Figure Extraction: PyMuPDF extracts embedded images from the PDF
3. Markdown Generation: Content is converted to Markdown with preserved structure
4. EPUB Creation: Markdown and images are transformed into EPUB3 with MathML/MathJax support

Why Nougat?

Nougat (Neural Optical Understanding for Academic Documents) is Meta's state-of-the-art model specifically designed for academic papers. It excels at:

• Recognizing complex mathematical notation
• Handling multi-column layouts
• Preserving table structures
• Extracting figures and captions

Model Sizes

Model	Size	Speed	Quality	Use Case
small	~350MB	Fast	Good	Quick conversions, testing
base	~1.2GB	Moderate	Better	Production use, complex papers

Performance

• CPU: 1-3 pages/minute (small model)
• GPU (CUDA): 10-20 pages/minute
• Apple Silicon (MPS): 5-15 pages/minute

Examples

Convert Multiple PDFs

for pdf in *.pdf; do
    paper2epub "$pdf" -a "Author Name"
done

Batch Processing in Python

from pathlib import Path
from paper2epub import Paper2EpubConverter

converter = Paper2EpubConverter()

pdf_dir = Path("papers")
for pdf_file in pdf_dir.glob("*.pdf"):
    print(f"Converting {pdf_file.name}...")
    converter.convert(pdf_file)

Limitations

• Scanned PDFs may require higher quality OCR (use base model)
• Very complex equations might need manual review
• Image quality depends on source PDF resolution
• EPUB readers vary in math rendering support (MathJax recommended)

Troubleshooting

Dependency Conflicts

Issue 1: albumentations

If you get an error about albumentations or ImageCompression:

# Install compatible version
pip install 'albumentations<1.4.0'

Issue 2: pypdfium2 (PdfDocument has no attribute 'render')

If you get an error about 'PdfDocument' object has no attribute 'render':

# Install compatible version
pip install 'pypdfium2>=4.0.0,<5.0.0'

Or reinstall with all fixes:

pip install --upgrade paper2epub

Out of Memory

# Reduce batch size
paper2epub paper.pdf -b 1

# Use CPU instead of GPU
paper2epub paper.pdf -d cpu

Poor Quality Output

# Use larger model
paper2epub paper.pdf -m base

# Enable verbose logging to debug
paper2epub paper.pdf -v

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE file for details.

Acknowledgments

• Nougat by Meta AI Research
• ebooklib for EPUB creation
• PyMuPDF for PDF handling

Citation

If you use paper2epub in academic work, please cite:

@software{paper2epub,
  title = {paper2epub: Academic PDF to EPUB Converter},
  author = {Komal Kumar},
  year = {2026},
  url = {https://github.com/MAXNORM8650/paper2epub}
}

For Nougat:

@article{blecher2023nougat,
  title={Nougat: Neural Optical Understanding for Academic Documents},
  author={Blecher, Lukas and Cucurull, Guillem and Scialom, Thomas and Stojnic, Robert},
  journal={arXiv preprint arXiv:2308.13418},
  year={2023}
}

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions

Roadmap

• GROBID integration for better metadata extraction
• Support for more input formats (DOCX, LaTeX)
• Batch processing UI
• Cloud/API deployment option
• Enhanced equation rendering options
• Custom styling templates