ome.zarr with labels not working

[dominikl]

I'm pretty stuck... I can't figure out why omero import returns "No imports found" as soon as you add a labels directory (Note: Tested with the 'labels' in filename issue fixed).

Here are four zarrs:

The "original" file, converted to zarr with bioformats which puts the image data into a "series" 0 directory:
tile_5.ome.zarr
-> Looks ok in validator, imports fine into omero.

The same zarr, but without the series 0 root directory:
tile_5_without_series.ome.zarr
-> Looks ok in validator, imports fine into omero.

The original zarr (with series '0') with labels (which are in a series '0' too):
tile_5_with_labels.ome.zarr
-> The validator sees the image in series '0', but not the labels. Doesn't import into omero.

0
0/0/...
0/1/...
...
labels
labels/Cells
labels/Cells/0
labels/Cells/0/0/...
labels/Cells/0/1/...
...

The zarr without series, with labels (now labels without series too):
tile_5_without_series_with_labels.ome.zarr
-> Looks ok in validator (image + labels). Doesn't import into omero.

0/...
1/...
...
labels
labels/Cells
labels/Cells/0/...
labels/Cells/1/...
...

Two questions:

• What actually is the correct 'with labels' format (series or not series)?
• If it's one of the two - Why does it not import?

Any idea? @will-moore @joshmoore @sbesson ?

[will-moore]

Without the "bioformats2raw" series, I would expect

image.ome.zarr/.zattrs        # image 'multiscales'
image.ome.zarr/0              
image.ome.zarr/0/.zarray  # first dataset resolution
image.ome.zarr/1              
image.ome.zarr/1/.zarray     # next dataset resolution
...
image.ome.zarr/labels/.zattrs       # contains  {"labels": ["Cells"] }
image.ome.zarr/labels/Cells/.zattrs       # contains 'multiscales'
image.ome.zarr/labels/Cells/0
image.ome.zarr/labels/Cells/0.zarray  # first dataset resolution

See example at https://ome.github.io/ome-ngff-validator/?source=https://uk1s3.embassy.ebi.ac.uk/idr/zarr/v0.4/idr0062A/6001240.zarr

E.g. https://uk1s3.embassy.ebi.ac.uk/idr/zarr/v0.4/idr0062A/6001240.zarr/labels/.zattrs is just:

{
    "labels": [
        "0"
    ]
}

I think that if it's nested under bioformats2raw layout, then every image.ome.zarr above would be image.ome.zarr/0.

[dominikl]

Yes, that's what I have in the files above too. Sorry, I just showed the basic directory structure without the .zattrs etc.
See tile_5_without_series_with_labels.ome.zarr: https://ome.github.io/ome-ngff-validator/?source=https://minio-dev.openmicroscopy.org/idr/with_labels/tile_5_without_series_with_labels.ome.zarr , resp. https://minio-dev.openmicroscopy.org/idr/with_labels/tile_5_without_series_with_labels.ome.zarr/labels/.zattrs . This should have exactly the same structure as your idr0062A example. I'll try that with "omero import", see what happens. Can I download that somehow (easily)?

[sbesson]

On the bioformats2raw question, we don t have any representative example (which is consistent with the fact that the tool only deals with images) but like for plates, my expectation is that the labels associated to a particular series should be nested under the relevant multiscales Zarr group i.e.

multiseries.zarr/
  0/                     # first series
    0/                   # first resolution of first series
    1/
    ...
    n/                   # last resolution of first series
    labels/           # labels of first series
      label1/         # label of first series
        0/
        ..
        n/
      label2/         # label of first series
        0/
        ..
        n/
  1/
    0/
    ...
    n/
    labels/
      ...
  OME/

Ideally the reader should be able to handle both layouts but we might indicate a preference and prioritize support for one of the two variants