docTR Labeler is a tool to label OCR data for the docTR and OnnxTR projects.

Attention: This project is still in development - and currently a pre-release version - please report any issues you encounter.

What you can expect from this repository:

• Efficient way to label OCR data
• Features like auto-annotation using OnnxTR and auto polygon adjustment
• Easy to use frontend with keybindings
• CLI and programmatic usage
• No Login required

Installation

Prerequisites

Python 3.10 (or higher) and pip are required to install docTR-Labeler.

Latest release

You can then install the latest release of the package using pypi as follows:

pip3 install doctr-labeler

Keybindings

• Ctrl + a : Select all polygons

• Esc : Deselect all selected polygons

• Ctrl + t : Auto adjust the selected polygons

• Ctrl + r : Reset last auto adjustment

• Ctrl + s : Save the current progress / image annotation

• Ctrl + d : Delete the selected polygon

• Ctrl + f : Draw a new polygon

• Ctrl + c : Undo while drawing a polygon

• Ctrl + + : Zoom in (up to 150% by default) - Can be changed by setting a environment variable DOCTR_LABELER_MAX_ZOOM to a value between 1.1 and 2.0

• Ctrl + - : Zoom out (down to 50% by default) - Can be changed by setting a environment variable DOCTR_LABELER_MIN_ZOOM to a value between 0.1 and 0.9

Configuration

You can set the following environment variables to configure the tool:

• DOCTR_LABELER_MAX_ZOOM : Maximum zoom level (default: 1.5)
• DOCTR_LABELER_MIN_ZOOM : Minimum zoom level (default: 0.5)
• RECOGNITION_ARCH : The recognition architecture to use (default: Felix92/onnxtr-parseq-multilingual-v1)
• DETECTION_ARCH : The detector architecture to use (default: fast_base)
• OBJECTNESS_THRESHOLD : The objectness threshold for the detector (default: 0.5)

Usage CLI

After installation you can use the CLI to start the tool:

For this open a terminal and run:

doctr-labeler

Usage Programmatic

You can also use the tool programmatic:

from labeler.views import GUI
from labeler.utils import prepare_data_folder, hf_upload_dataset

# (Optional)
# Prepare the data folder you can pass a path to a folder containing images and PDFs
# The function will create a new folder 'images' with the prepared data
prepared_data_path = prepare_data_folder("path/to/folder")

# Start the GUI
gui = GUI(image_folder=prepared_data_path)
gui.start_gui()

# or if you want to annotate also for KIE
types = ["Total", "Date", "Invoice Number", "VAT Number", "Address", "Company Name"]
gui = GUI(image_folder=prepared_data_path, text_types=types)
gui.start_gui()

# (Optional) Upload the prepared data to the Hugging Face dataset hub
# The path to the folder should contain an 'images' folder and it's corresponding 'labels.json' file or the 'tmp
hf_upload_dataset(prepared_data_path)

Credits

• This project is based on the Form-Labeller project by Devarshi Aggarwal.

Citation

If you wish to cite please refer to the base project citation, feel free to use this BibTeX references:

@misc{docTR-Labeler,
    title={docTR Labeler: docTR OCR Annotation Tool},
    author={{Dittrich, Felix}, {List, Ian}},
    year={2024},
    publisher = {GitHub},
    howpublished = {\url{https://github.com/text2knowledge/docTR-Labeler}}
}

@misc{Form-Labeller,
  author = {Aggarwal, Devarshi},
  title = {{Form Labeller}},
  howpublished = {\url{https://github.com/devarshi16/Form-Labeller}},
  year = {2020},
  note = {Online; accessed 01-March-2020}
}

Contributing

Contributions are what make the open-source community such an amazing place to learn, inspire, and create.

Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Add your Changes
4. Run the tests and quality checks (make test and make style and make quality)
5. Commit your Changes (git commit -m 'Add some AmazingFeature')
6. Push to the Branch (git push origin feature/AmazingFeature)

License

Distributed under the Apache 2.0 License. See LICENSE for more information.