Make image label key mandatory for label images

[dstansby]

Fixes ome/ngff#339 - see discussion on that thread for motivation.

[github-actions]

Automated Review URLs

• Readthedocs

[jo-mueller]

Potential duplicate/follow-up to ome/ngff#170

[jo-mueller]

@dstansby sorry that I haven't reacted here earlier. I'm a bit uncertain about approving this (or not) because I'm not sure about the implications. I.e., I know why this is helpful, but it would also be useful to know why the requirement was weakened from a MUST to a SHOULD in the first place. I.e., if there's a reason that we're not aware of that led to changing this, it would be good to know so we don't repeat potential mistakes of the past 😬

[dstansby]

As far as I can see the change was made in ome/ngff#165. The only discussion of the image-label key being mandatory or not there was a single comment at ome/ngff#165 (comment), which was not picked up or discussed any further. I'm therefore not really sure what the context/motviation was for making this change in the first place - perhaps @virginiascarlett and @bogovicj have throughts as the authors of ome/ngff#165?

[jo-mueller]

Looking at previous version, the statement was usually phrased like this:

Groups containing the image-label dictionary represent an image segmentation in which each unique pixel value represents a separate segmented object.

Which to me reads more like a MUST than a SHOULD but it would be nice if others could weigh in here a bit (discussion for next techdev meeting?)

[dstansby]

👍 to discuss at a meeting, but since that's not a public meeting I thnk folks should read and add their thoughts to ome/ngff#339 so all the context and discussion is in one place instead of fragmented.

[dstansby]

Does anyone from the OME side have thoughts on the next steps here? It would be nice to have a decision on whether this is merged in time for 0.6.

[jo-mueller]

Where I am concerned, I'll say 👎 Long story short, I think a label image should not be flagged as an invalid image only because it doesn't declare itself to be a label image properly. That will still mean, that it is a valid image (although the default color settings viewers may choose may not make much sense for it).

A longer version of this line of thought is found in this issue. Making this declaration more explicit or introduce a sort of typing of images is something I see more on the plate of RFC8, if anywhere.

[dstansby]

a label image should not be flagged as an invalid image only because it doesn't declare itself to be a label image properly.

Without the image-label key, the data will still consitute a valid OME-Zarr image group. So I'm not sure why this is a concern?

[jo-mueller]

I would read

In addition to the multiscales key, the OME-Zarr Metadata in this image-level zarr.json file MUST contain another key, image-label,

as an equivalent statement to "If it is a label image and doesn't have the image-label field present, it is invalid?

[dstansby]

If it is a label image and doesn't have the image-label field present, it would be an invalid image label group, but that doesn't prevent it from being a valid image group.