image_io.single_position_to_zarr sets the chunk and shape value for any of the default axes (time, channel, z, y, x) that's not present in the structure of the shape attribute of the given image to 1. But this means that the value of the chunks and of the shape key in the zarr metadata can be incorrect, alleging that the underlying data array has dimensions which aren't really there.
This can be hacked around by doing single-value slice indexing into the underlying numpy array so that any singleton dimension is preserved, but we should have a cleaner solution, in which we don't meddle with the actual structure of the data in order to have it conform to our metadata specification.
vreuter
commented
5 months ago
image_io.single_position_to_zarrsets the chunk and shape value for any of the default axes (time, channel, z, y, x) that's not present in the structure of theshapeattribute of the given image to 1. But this means that the value of thechunksand of theshapekey in the zarr metadata can be incorrect, alleging that the underlying data array has dimensions which aren't really there.Requirement for
chunksandshape: https://zarr.readthedocs.io/en/stable/spec/v2.htmlThis can be hacked around by doing single-value slice indexing into the underlying numpy array so that any singleton dimension is preserved, but we should have a cleaner solution, in which we don't meddle with the actual structure of the data in order to have it conform to our metadata specification.