Missing models to upload

[esgomezm]

Hi @crtorreg and @xion16lm

I'm reviewing which are the models that still need to be updated/uploaded to the bioimage model zoo using the uploader. Here I let you the list from what I saw. Could you please take a look and update them?

• UNet2DGlioblastomaSegmentation Needs to be uploaded and the download link updated with the new one https://github.com/deepimagej/models/blob/8c20e945b0e8e5201fe34957ab470dcf35f92e94/manifest.bioimage.io.yaml#L189

• SMLMDensityMapEstimationDEFCoN The download link needs to be updated with the new one. @oeway you have uploaded this model ( https://bioimage.io/#/?type=all&tags=density&id=10.5072%2Fzenodo.849032) but we need to add a zip file so we can link it to the download field. Could one of us send you the updated version of this model so you can update it in the bioimageio?

• DeepSTORMZeroCostDL4Mic The download link needs to be updated.

• 2DUNetZeroCostDL4Mic The download link needs to be updated.

• JonesVirtualStaining The download link needs to be updated.

• Mt3VirtualStaining The download link needs to be updated.

• Usiigaci The download link needs to be updated

• SkinLesionClassification The download link needs to be upddated

• 3DUNetZeroCostDL4Mic The download link needs to be updated

Note that the basic package we need to upload for deepImageJ models (until we can call the bioengine packager) is the following one:

• All the files specified in the BioImage Model Zoo
• A zip file containing the bioimage model zoo model compatible with deepImage. Before submitting it, we need to verify that the model runs in deepImageJ. The final model.yaml needs to be updated in this repository. Once all the model items are uploaded in zenodo, they can be removed from this repository.

Now there is the spec checker for the uploads and most probably none of our models passes the spec because we are missing the batch dimensions and the channel dimension (when it is one) in all the numpy arrays. So, if you are uploading a new model (like UNet2DGlioblastomaSegmentation), please change this.

@oeway:

• is there any field in the rdf in which we can specify the id of the model so deepImageJ can directly call to it in the future using the bioengine? or is it enough by nowing the id (e.g. deepimagej/SkinLesionClassigication)?
• should we also change the aplications and datasets from this yaml to the regular upload of the bioimage model zoo? (this is to stop using the yaml to render the bioimage.io as we already spoke).

[crtorreg]

Hi @esgomezm,

I am going to take a look one by one now!

[esgomezm]

OK! Thank you @crtorreg !

Please, note that whenever we make a new commit, GitHub workflows check that the model specs are correct. If this workflow does not pass the test (red cross in the image), then, the json file is not updated and therefore, we won't see the changes in the deepImageJ installer:

At the moment the validator has some errors (see issue here), but once this is solve, please check that your changes are correct.