Add API for writing image-label

[constantinpape]

Is there support to write "image-label" datasets (https://ngff.openmicroscopy.org/latest/#label-md) in ome-zarr-py already?
If yes, how do I use it?

[will-moore]

Threre's an example at https://github.com/ome/ome-zarr-py/blob/0ea5bf61ef754069abf4f223897d55630c953fbf/README.rst
which is part of a long-standing PR #121
Actually, I have subsequently removed the labels from the PR to make it more concise for the README

[constantinpape]

Thanks @will-moore; the example was very helpful and I could update my scripts to write ome.zarr with labels.

[sbesson]

Looking at the example, I could imagine wrapping the succession of low-level calls into utilities dealing with the group/metadata creation/update e.g.

write_image(image=data, group=image_root)
label_root = add_label(image_root, label_name)
write_image(image=label_data, group=label_root)

or

write_image(image=data, group=image_root)
write_label(image=label_image, group=image_root, name=label_name)

It would be interesting to know you would see the value of introducing such API based on your usage @constantinpape.

[constantinpape]

It would be interesting to know you would see the value of introducing such API based on your usage @constantinpape.

Yes, absolutely! write_label should also have support for (optional? depending on ome/ngff#105) arguments to pass the label/property values.

[sbesson]

I have updated the title of this issue to reflect the new scope.

So current preference would be rather a single API that would first create/update the intermediate groups (depending on the presence of labels) and then write the label image, is that correct? In terms of behavior, I agree the API should also support both mandatory as well as optional image-label attributes.

For the metadata, my current preference would be to extend write_multiscale to support an extra metadata dictionary similarly storage_options. This would allow to set additional metadata, either specified (omero, image-label...) or custom, to be stored in the attributes of the multiscales group.

Thinking of the sequence of a write_label_image API, the biggest unknown to me at this point is whether this API should call write_image or write_multiscales. The former is effectively a wrapper around the latter but I do not know whether it should always be seen as the single entry-point @joshmoore @will-moore ?

[will-moore]

I think users will find write_image() and write_multiscale() useful, since they may have their pyramid already generated (not want to use the Scaler).

[constantinpape]

As discussed in ome/napari-ome-zarr#33 this API would need to also add the group-level "labels" metadata, which btw should be explained better in the spec, see ome/ngff#108.

[constantinpape]

Implemented in #178