Update "stardist" model and datasource for scivision catalog

[edwardchalstrey1]

Hi @mooniean - I am in the process of updating scivision model/datasource entries for the catalog/website, can you help me with the following?

1. I assume as this is cell bio you know the authors? If you can't answer the below yourself, could you tag the GH account of the contact who we got this from?
2. Can you confirm the model and dataset thumbnail images are appropriate
3. Can you confirm the descriptions are appropriate?
4. Is there a better name we could give the datasource?
5. Are there any scivision notebooks where stardist data or model are loaded that would break as a result of these name changes? I assume not, since the model is not currently loadable via the Python package.

TODO:

• Looks like the model might be too complicated to make installable via scivision Python package, but we could set up the dataset with a .scivision/data.yml (so we can use load_dataset) - created a new issue: Make stardist datasource loadable via Python api #482

See the screenshots below for what this looks like in Scivision

[edwardchalstrey1]

Scivision model page

Scivision datasources page

[ots22]

Currently the catalogs have a few older entries like this that point to repositories that aren’t set up as ‘scivision compatible’ models/datasets (without .scivision/ metadata).

We could write wrappers around these, which is probably the ideal outcome, but what should we do before that?

We could

1. Leave them in the catalog
   • perhaps add a ‘has-scivision-metadata’ field to the catalog, indicating that they can’t be run or loaded automatically - or maybe just detecting the missing .scivision is enough. These could be visible in the frontend, but fail in the python library with a sensible message.
2. Remove them for now
   • add them to a list of models/datasets we want/intend to include somewhere (perhaps 'good first issue' material)

(Similar slack discussion)

[edwardchalstrey1]

Currently the catalogs have a few older entries like this that point to repositories that aren’t set up as ‘scivision compatible’ models/datasets (without .scivision/ metadata).

We could write wrappers around these, which is probably the ideal outcome, but what should we do before that?

We could

1. Leave them in the catalog

   • perhaps add a ‘has-scivision-metadata’ field to the catalog, indicating that they can’t be run or loaded automatically - or maybe just detecting the missing .scivision is enough. These could be visible in the frontend, but fail in the python library with a sensible message.

2. Remove them for now

   • add them to a list of models/datasets we want/intend to include somewhere (perhaps 'good first issue' material)

(Similar slack discussion)

yeah, in the frontend we could have a tag like scivision.py loadable

[mooniean]

Somehow I missed the notification for this - I actually haven't worked with this, I believe it was @evangeline-corcoran

[edwardchalstrey1]

Hi @evangeline-corcoran when you have a free moment would you be able to take a look at the questions I have listed in the top comment, no rush :)

[evangeline-corcoran]

Hi @edwardchalstrey1

Please see my answers to the top comment questions below and let me know if you need any more info

> I assume as this is cell bio you know the authors? If you can't answer the below yourself, could you tag the GH account of the contact who we got this from?

The authors are Uwe Schmidt (uschmidt83), Martin Weigert (maweigert), Robert Haase (haesleinhuepf), and Ko Sugawara (ksugar). The stardist models and datasets are publicly available here: https://github.com/stardist/stardist

> Can you confirm the model and dataset thumbnail images are appropriate

Yes, these are good

> Can you confirm the descriptions are appropriate?

For the model I would suggest - "Single class object detection and segementation of star-convex polygons"

For the dataset - "Cell microscopy test dataset for detection and segmentation of individual nuclei"

> Is there a better name we could give the datasource?

Something like "StarDist cell nuclei 2D" as they also have a dataset of synthetic 3D cell images

> Are there any scivision notebooks where stardist data or model are loaded that would break as a result of these name changes? I assume not, since the model is not currently loadable via the Python package.

No

[edwardchalstrey1]

Hi @uschmidt83 @maweigert @haesleinhuepf @ksugar - we're currently in the process of upgrading the Scivision project and included datasets/models with a new website available at https://sci.vision - as a result of this we're updating some of the metadata for legacy models/datasources including your own that were added at an earlier stage of the project. Would one of you be able to review the proposed names, descriptions and thumbnails in this PR and let us know your thoughts?

In addition, we aim wherever possible to make inclusions to the scivision catalog loadable via the Python API. I can make this possible for your datasource, however I notice at the moment that the datasource url (https://github.com/stardist/stardist/releases/download/0.3.0/demo3D.zip) downloads a folder of both training and test images (and masks), but that the images appear to be blank? Perhaps I'm misunderstanding something about this data, but I was expecting images similar to thumbnail I added (I think taken from the model GitHub page). If possible, a folder of test images that are ready for the model to be run on would be good for inclusion in scivision.

Let me know your thoughts (and no rush on this). Thanks, Ed.

[edwardchalstrey1]

I'm going to merge this PR now to get the thumbnail images in, but we can always change things at a later date if needed