Python module for data labeling leveraging the Universal Data Tool.
- Open Universal Data Tool in Jupyter notebook
- Massage data into and out of the UDT format
pip install universaldatatoolimport universaldatatool as udt
ds = udt.Dataset(
type="image_segmentation",
image_paths=["/path/to/birds/good_bird.jpg","/path/to/birds/bird2.jpg"],
labels=["good bird", "bad bird"]
)
# Opens dataset directly in jupyter notebook
ds.open()- udt.nb: jupyter notebook widget
- udt.load_json(file_path): Load UDT File from json
- udt.load_csv(file_path): Load UDT File from csv
- udt.Dataset(type=None, image_paths=None, labels=None)
- udt.Interface(type=None, labels=None): Create UDT interface
- udt.Sample(image_url=None, document=None, ...) : Create UDT Sample
- udt.nb.display(udt_file): Display Universal Data Tool widget
-
image_path,video_pathetc. support - Better Docs
- Usage Examples
- Load CSV or JSON from files
- Collaborative synchronization w/ universaldatatool.com
-
edit/openshould check that there are no local paths - Helpful stringification
- Make it easy to run tests
- Image Segmentation kills jupyter notebook scrolling
- Make JupyterLab Extension 1 2 3
- Continuous integration testing via Github Actions
- Cypress browser testing
Cypress will automatically open a browser and create jupyter notebooks with different test scenarios. It's really fast for developing and testing. To use it, you must first run our jupyter docker container, which mounts volumes properly such that universaldatatool can be imported. To do this, run:
yarn start:jupyterA jupyter notebook is now running in the background.
You can now run the cypress tests in development mode by running...
yarn cy:runAn electron browser will open with automated tests.
Each file in the universaldatatool/tests directory can be tested with pytest e.g.
python -m pytest universaldatatool/tests/example1.py