ZarrReader doesn't read zarrs without top-level 'series'

[dominikl]

Bioformats2raw creates ome.zarrs with "0" top-level directory ("series" : [ "0" ] in OME/.zattr). These import fine. But when you remove this top level directory, it reports ("no imports found"):

(venv3) bash-5.1$ /opt/omero/server/OMERO.server/bin/omero import -T Dataset:18451 --transfer=ln_s --depth=100 -n /uod/idr/filesets/idr0138-lohoff-seqfish/20210917-Globus/ngff/TimEmbryos-102219/HybCycle_0/MMStack_Pos0.ome.zarr
Using session for demo@localhost:4064. Idle timeout: 10 min. Current group: Public
2024-10-28 14:54:58,602 313        [      main] INFO          ome.formats.importer.ImportConfig - OMERO.blitz Version: 5.7.4
2024-10-28 14:54:58,620 331        [      main] INFO          ome.formats.importer.ImportConfig - Bioformats version: 7.3.1 revision: 41e12f630ad7349caa0293cba23948c17d22732b date: 11 July 2024
2024-10-28 14:54:58,678 389        [      main] INFO   formats.importer.cli.CommandLineImporter - Setting transfer to ln_s
2024-10-28 14:54:58,682 393        [      main] INFO   formats.importer.cli.CommandLineImporter - Log levels -- Bio-Formats: ERROR OMERO.importer: INFO
2024-10-28 14:54:59,073 784        [      main] INFO      ome.formats.importer.ImportCandidates - Depth: 100 Metadata Level: MINIMUM
No imports found

I don't think this top-level series is mandatory, is it? At least for older ome.zarrs (see above example from idr0138) we didn't use that. A previous version of the ZarrReader must have supported that.

[dominikl]

/cc @sbesson any ideas? The currently imported idr0138 images are fine though. So ZarrReader can read the image data without the 0 series. But for some reason for the import it expects the 0 series top-level directory.

[sbesson]

I suspect whatever changes caused the issue were introduced as part of the overhaul in #64.

What do you mean by "Remove this top level directory"? Are you referring to the OME Zarr group or the 0 Zarr group?

[dominikl]

Ah, that's possible. I see if I can fix that...

With removing I mean move the contents out of this '0' top-level directory, one level up (tile_5.ome.zarr vs tile_5_with_labels.ome.zarr):

And remove the '0' series from OME/.zattrs.

{
  "series" : [ "0" ]
}

The resulting zarr looks fine imo, according to validator, and can also be viewed in vizarr. But unfortunately not imported (no imports found).

https://ome.github.io/ome-ngff-validator/?source=https://minio-dev.openmicroscopy.org/idr/with_labels/tile_5_with_labels.ome.zarr

[sbesson]

Is OME/METADATA.ome.xml present in your directory? The reader definitely includes some logic which parses this XML and tries to map it into the Zarr groups. If that's the source of the issue, you might look into tightening the conditions
of detection of this file

[dominikl]

Same problem after I removed OME directory. Also idr0138 don't have OME dir and show the same "no imports" issue. Anyway, I'll check the PR you mentioned, maybe I can find out where this 0 series assumption sneaked in.

[dominikl]

So for my ...with_labels.ome.zarr example the problem is actually the "labels" in the filename! Fixed here: b2c2b9c . Doesn't explain the "no imports" for idr0138 zarrs. But I'll look into that again later.