AllenCellModeling
Default tiff reader missing planes (with specific file)
System and Software
- aicsimageio Version: 4.9.2
- Python Version: 3.10.5
- Operating System: macos
Description
Not sure what it is about this file, and I haven't looked too deeply at the metadata yet, but the default TiffReader class seems to miss the z planes in the linked file, whereas the BioformatsReader and tiffile both see it:
https://www.dropbox.com/s/ls6ci1cxyocqbgr/2MF1P2_glia.tif?dl=0
In [28]: im = AICSImage('2MF1P2_glia.tif')
In [29]: bim = AICSImage('2MF1P2_glia.tif', reader=BioformatsReader)
In [30]: tim = tf.imread('2MF1P2_glia.tif')
In [31]: im.shape
Out[31]: (1, 1, 1, 1024, 1024)
In [32]: bim.shape
Out[32]: (343, 1, 1, 1024, 1024)
In [33]: tim.shape
Out[33]: (343, 1024, 1024)
and to confirm, they're also not detected as scenes:
In [38]: im.scenes
Out[38]: ('Image:0',)
Expected Behavior
I expected a shape of 343 planes
Interesting case. I think (1, 1, 1, 1024, 1024) might be the expected shape for dimension order 'TCZYX'. The file 2MF1P2_glia.tif was written by SCIFIO. It uses a provisional, undocumented metadata format, which is not supported by Bio-Formats or tifffile:
SCIFIO=0.42.0
axes=X,Y,Unknown
lengths=1024,1024,343
scales=1.0,1.0,1.0
units=null,null,null
bitsPerPixel=8
images=343
channels=1
slices=1
frames=1
hyperstack=true
mode=composite
unit=null
Bio-Formats handles the unknown axis as time, while tifffile uses the generic axis "I", which is pruned by aicsimageio...
I haven't dived into the code around this yet, but aicsimageio definitely wants to put the planes with their proper semantic dimension, with as little guessing as possible. That's a big reason why "I" gets pruned.
Maybe there is a way for us to say if slices=1 (Z) and channels=1 (C) then the only thing left is T. Speculating because the implementation wraps tifffile so those may not be exposed in the way i'm suggesting.
