docling_ocr

A powerful Python package for extracting text from images and documents using advanced LLM-based models.

Overview

docling_ocr leverages state-of-the-art language models specifically designed for document understanding tasks. Unlike traditional OCR engines that rely solely on character recognition, docling_ocr uses language models that understand document context, layouts, and can handle various document formats with high accuracy.

Built on top of models like SmolDocling, this package provides a simple, intuitive interface for document text extraction tasks.

Features

• LLM-powered extraction: Uses advanced language models trained specifically for document understanding
• Context-aware recognition: Understands document layouts and context for improved accuracy
• Multi-format support: Works with scanned documents, forms, receipts, and other text-heavy images
• Simple API: Easy-to-use interface with both file and image object inputs
• Batch processing: Process entire directories of documents efficiently
• Flexible output options: Return text or save directly to files
• Extensible architecture: Abstract base class makes it easy to add new models

Installation

pip install docling_ocr

Requirements

• Python 3.7+
• PyTorch 1.10.0+
• Transformers 4.15.0+
• Pillow 8.0.0+

Quick Start

Basic Usage

from docling_ocr import SmolDoclingExtractor

# Initialize the extractor
extractor = SmolDoclingExtractor()

# Extract text from an image file
text = extractor.extract_text("path/to/document.jpg")
print(text)

# Or use the shorthand callable interface
text = extractor("path/to/document.jpg")

Using with PIL Images

from docling_ocr import SmolDoclingExtractor
from PIL import Image

# Initialize the extractor
extractor = SmolDoclingExtractor()

# Open image with PIL
image = Image.open("path/to/document.jpg")

# Extract text
text = extractor.extract_text_from_image(image)
print(text)

Batch Processing

from docling_ocr import SmolDoclingExtractor
from docling_ocr.utils import batch_process

# Initialize extractor
extractor = SmolDoclingExtractor()

# Process all images in a directory
results = batch_process(
    extractor, 
    image_dir="path/to/documents/", 
    output_dir="path/to/output/",
    extensions=['.jpg', '.png', '.pdf']  # Optional: specify file extensions
)

# Results contains a dictionary mapping filenames to extracted text
for filename, text in results.items():
    print(f"File: {filename}")
    print(f"Text: {text[:100]}...")  # Print first 100 chars
    print("-" * 50)

Advanced Usage

GPU Acceleration

By default, the extractor will use CUDA if available. You can explicitly specify the device:

# Use CPU explicitly
extractor = SmolDoclingExtractor(device="cpu")

# Use specific GPU
extractor = SmolDoclingExtractor(device="cuda:0")

Custom Model Configuration

You can specify a different model from the same family:

# Use a different model variant
extractor = SmolDoclingExtractor(model_name="ds4sd/SmolDocling-512M")

Adjusting Generated Text Length

For longer documents, you may want to increase the maximum generated text length:

# Extract with a longer maximum length for complex documents
text = extractor.extract_text("complex_document.pdf", max_length=1024)

Performance Considerations

• Processing time depends on the image size, complexity, and hardware
• GPU acceleration is recommended for batch processing
• First initialization loads the model which may take some time
• Subsequent calls are much faster as the model remains in memory

Comparison with Traditional OCR

docling_ocr differs from traditional OCR engines in several key ways:

Feature	Traditional OCR	docling_ocr
Text Recognition	Character/word based	Context-aware language understanding
Layout Understanding	Limited/separate process	Integrated understanding
Language Understanding	Limited	Leverages LLM language capabilities
Format Flexibility	Engine-specific	Adaptable to various formats
Context Retention	Limited	Maintains document context

Examples

Forms and Structured Documents

from docling_ocr import SmolDoclingExtractor

extractor = SmolDoclingExtractor()
form_text = extractor("tax_form.jpg")
print(form_text)

Tables and Spreadsheets

spreadsheet_text = extractor("financial_data.jpg")
print(spreadsheet_text)

Receipts and Invoices

receipt_text = extractor("receipt.jpg")
print(receipt_text)

Contributing

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

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add some amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Future Roadmap

• Support for PDF documents with multi-page handling
• Additional LLM-based extraction models
• Fine-tuning options for specific document types
• Structured data extraction (JSON output)
• Layout-preserving extraction options

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built on the amazing work of the SmolDocling team for the SmolDocling-256M-preview model.
• Inspired by the growing field of document AI
• Thanks to the HuggingFace team for making transformers accessible

Citation

If you use this package in your research, please cite:

@software{docling_ocr,
  author = {Adhing'a Fredrick},
  title = {docling_ocr: LLM-based Document Text Extraction},
  year = {2025},
  url = {https://github.com/FREDERICO23/docling_ocr}
}

Contact

For questions and support, please open an issue on the GitHub repository or contact [email protected].