BioImage.IO Model Manager
napari plugin for managing AI models in the BioImage Model Zoo.
WARNING: This is an alpha release. The API may change in future versions, and please feel free to create issues to report bugs or provide feedbacks.
pip install napari-bioimageio
(If you don't have napari installed, run
pip install napari[pyqt5])
This library is meant for helping developers to ease the handling of models in napari.
We provide a set of API functions for managing and selecting models.
Show the model manager with a model list pulled from the BioImage Model Zoo, the user can explore all the available models, download or remove models.
Display a dialog for selecting models from the BioImage Model Zoo, the user can either select an existing model or download from the BioImage Model Zoo.
The selecte model information (a dictionary) will be returned if the user selected a model, otherwise it returns
Once the user selected the model, you can access the name, and also the file path to the model resource description file (via the
rdf_source key). With the
bioimageio.core library (installed via
pip install bioimageio.core or
conda install -c conda-forge bioimageio.core), you can run inference directly, the following examples shows how to implement it:
# Popup a model selection dialog for choosing the model model_info = show_model_selector(filter=nuclear_segmentation_model_filter) if model_info: self.nucseg_model_source = model_info["rdf_source"] # Load model model_description = bioimageio.core.load_resource_description(model_info["rdf_source"]) input_image = imageio.imread("./my-image.tif") with bioimageio.core.create_prediction_pipeline( bioimageio_model=model_description ) as pipeline: output_image = bioimageio.core.prediction.predict_with_padding( pipeline, input_image, padding=padding )
Note: To run the models, you need to setup the conda environment properly according to the installation guide of bioimageio.core.
For more examples, see this example notebook for
You can also access the weight files directly by searching the model folder (e.g. extract the model folder path via
os.path.dirname(model_description["rdf_source"])), this will be useful if you prefer to use your own model inference logic.
Display a dialog to instruct the user to upload a model package to the BioImage Model Zoo. Currently, it only shows a message, in the future, we will try to support direct uploading with user's credentials obtained from Zenodo (a public data repository used by the BioImage Model Zoo to store models).
To create a BioImageIO-compatible model package, you can use the
build_model function as demonstrated in this notebook.
Install and set up development environment.
pip install -r requirements_dev.txt
This will install all requirements. It will also install this package in development mode, so that code changes are applied immediately without reinstall necessary.
Here's a list of development tools we use.
It's recommended to use the corresponding code formatter and linters also in your code editor to get instant feedback. A popular editor that can do this is
Run all tests, check formatting and linting.
Run a single tox environment.
tox -e lint
Reinstall all tox environments.
Run pytest and all tests.
Run pytest and calculate coverage for the package.
pytest --cov-report term-missing --cov=napari-bioimageio
Continuous integration is by default supported via GitHub actions. GitHub actions is free for public repositories and comes with 2000 free Ubuntu build minutes per month for private repositories.
- 07 July 2022
- 05 July 2022
- Information not submitted
- Information not submitted
- Stars: 0
- Forks: 0
- Issues + PRs: 0