Version v.1.0 includes a major refactoring. Key changes include:
- PyTorch-only support for all deep learning models.
- Support for many more fine-tuned models from the Huggingface Hub (Bert, RobertA, LayoutLM, LiLT, ...)
- Decomposition into small sub-packages: dd-core, dd-datasets and deepdoctection
- Type validations of core data structures
- New test suite
deepdoctection is a Python library that orchestrates Scan and PDF document layout analysis, OCR and document and token classification. Build and run a pipeline for your document extraction tasks, develop your own document extraction workflow, fine-tune pre-trained models and use them seamlessly for inference.
- Document layout analysis and table recognition in PyTorch with Detectron2 and Transformers,
- OCR with support of Tesseract, DocTr and AWS Textract,
- Document and token classification with the LayoutLM family, LiLT and and many Bert-style models including features like sliding windows.
- Text mining for native PDFs with pdfplumber,
- Language detection with with transformer based
papluca/xlm-roberta-base-language-detection. - Deskewing and rotating images with jdeskew or Tesseract.
- Fine-tuning object detection, document or token classification models and evaluating whole pipelines.
- Lot's of tutorials
Have a look at the introduction notebook for an easy start.
Check the release notes for recent updates.
Check the demo of a document layout analysis pipeline with OCR on 🤗 Hugging Face spaces.
The following example shows how to use the built-in analyzer to decompose a PDF document into its layout structures.
import deepdoctection as dd
from IPython.core.display import HTML
from matplotlib import pyplot as plt
analyzer = dd.get_dd_analyzer() # instantiate the built-in analyzer similar to the Hugging Face space demo
df = analyzer.analyze(path = "/path/to/your/doc.pdf") # setting up pipeline
df.reset_state() # Trigger some initialization
doc = iter(df)
page = next(doc)
image = page.viz(show_figures=True, show_residual_layouts=True)
plt.figure(figsize = (25,17))
plt.axis('off')
plt.imshow(image)
HTML(page.tables[0].html)
print(page.text)
- Python >= 3.10
- PyTorch >= 2.6
- To fine-tune models, a GPU is recommended.
We recommend using a virtual environment.
For a simple setup which is enough to parse documents with the default setting, install the following
uv pip install timm # needed for the default setup
uv pip install transformers
uv pip install python-doctr
uv pip install deepdoctection
This setup is sufficient to run the introduction notebook.
The following installation will give you a general setup so that you can experiment with various configurations. Remember, that you always have to install PyTorch separately.
First install Detectron2 separately as it is not distributed via PyPi. Check the instruction here or try:
uv pip install --no-build-isolation detectron2@git+https://github.com/deepdoctection/detectron2.git
Then install deepdoctection with all its dependencies:
uv pip install deepdoctection[full]
For further information, please consult the full installation instructions.
Download the repository or clone via
git clone https://github.com/deepdoctection/deepdoctection.git
The easiest way is to install with make. A virtual environment is required
make install-ddPre-existing Docker images can be downloaded from the Docker hub.
Additionally, specify a working directory to mount files to be processed into the container.
docker compose up -d
will start the container. There is no endpoint exposed, though.
We thank all libraries that provide high quality code and pre-trained models. Without, it would have been impossible to develop this framework.
...you can easily support the project by making it more visible. Leaving a star or a recommendation will help.
Distributed under the Apache 2.0 License. Check LICENSE for additional information.