Note: This website is under construction.

An AI model zoo for COVID-19 related research

A list of models can be found here.

COVID-19 model zoo is a static website, it reads a manifest file from this Github repo (manifest.model.json) and render the page in the browser. Therefore, contributing models to the repository is as easy as adding links of your models to the manifest file.

Please follow the following steps:

1. Fork this repo
2. Add your models to the src/manifest.model.yaml file
3. Run python src/compile_model_manifest.py to generate a new manifest.model.json with your models
4. Commit your changes and push to your Github repo.
5. You can preview it constructing an URL which makes COVID-19 model zoo render the page with the manifest file in your repo. The URL format is: https://covid-ai.io?repo=YOUR_GITHUB_USER_NAME/YOUR_GITHUB_REPO, for example: https://covid-ai.io?repo=oeway/models will point to the model manifest hosted on https://github.com/oeway/models. You can also add commit hash tag, branch name or tag after that, for example: https://covid-ai.io?repo=oeway/models/06a9ffac88.
6. If you are satisfied with the result above, you can send us a Pull Request, and we will review it before it get merged.

Each model in the BioImage model zoo can associated with a list of applications which you can run directly by the end user. We use BioEngine, a tailored version of ImJoy to run these applications. Therefore, you can basically run ImJoy plugins with the BioEngine specific api. By default it loads also a Jupyter Engine which uses free computational resources on MyBinder.org, so you can also run small models in Python.

Since BioEngine is designed for running model specific ImJoy plugins, it needs to define either runOneModel() and/or runManyModels() function in the plugin api. Plus, you need also a testModel function which will be used to run tests in a CI environment. For example, the following python plugin would treat as a qualified BioEngine App:

from imjoy import api

class ImJoyPlugin():
    def setup(self):
        pass

    def run(self, ctx):
        pass

    def runOneModel(self, model_info):
        # run the model
        pass

    def runManyModels(self, model_info_list):
        # filter the model list and run them
        pass
    
    def testModel(self):
        # run tests for the model
        pass

api.export(ImJoyPlugin())

Note, this means your plugin would contain 4~5 functions including setup and run which is required for ImJoy.

You don't need to return any value after execution. In case of error, you can just throw or raise the error.

You can do the debugging inside ImJoy, for more information, please consult https://imjoy.io/docs.

To test with the BioEngine, you can go to https://covid-ai.io, on the menu located in the top-right corner, you can load a local ImJoy plugin file to run it with the BioEngine. One additional feature is that the BioEngine will keep track of the local file, if you made new changes with your code editor (e.g. vim, vscode) the engine will try to reload the plugin file.

You can submit your BioEngine App by changing the same file named src/manifest.model.yaml as for contributing models.

Here are the steps:

1. Once the BioEngine App is ready, you can then push it to your Github repo and get a raw URL for the file.
2. Define a key in the applications section in src/manifest.model.yaml, and set the value as the raw URL to the BioEngine app file.
3. For all the models which your app can digest, you can add your app key to the applications field of the model.
4. The procedure later are the same as contributing models, you can basically: run python src/compile_model_manifest.py to generate a new manifest.model.json, commit and push to your Github repo, preview it on covid-ai.io with https://covid-ai.io?repo=YOUR_GITHUB_USER_NAME/YOUR_GITHUB_REPO and optionally send us a Pull Request.

If you are developing or maintaining a software which can digest models in the model description format used in the BioImage model zoo, it's easy to fetch the latest model list (along with the other meta information) from this url https://raw.githubusercontent.com/CellProfiling/covid19ai/master/manifest.model.json.

By fetching the content of this JSON file, you will get all the models from the BioImage model zoo, and you can filter it based on your own supported framework, programming language etc.