Restrict units to what is defined in NGFF schema #289
Conversation
There was a problem hiding this comment.
Using bioformats2raw 0.10.1, I was able to reproduce the original issue either using the Fiji workflow described in #288 (Open the MRI sample and export as OME-TIFF) or a synthetic sample file like test&physicalSizeX=1pixel&physicalSizeY=1pixel.fake. In both cases, the conversion generates a Zarr dataset where the unit of the x and y the physical axes are expressed in pixel.
With a build of bioformats2raw including this PR, the Zarr datasets generated by the conversion are binary identical except for the physical axes which are now unitless
sbesson@Sebastien-GS-MacBook-Pro-2025 bioformats2raw % diff -ur test.zarr test_pr289.zarr
diff -ur test.zarr/0/.zattrs test_pr289.zarr/0/.zattrs
--- test.zarr/0/.zattrs 2025-09-04 08:46:45
+++ test_pr289.zarr/0/.zattrs 2025-09-04 08:54:26
@@ -14,11 +14,9 @@
"name" : "z",
"type" : "space"
}, {
- "unit" : "pixel",
"name" : "y",
"type" : "space"
}, {
- "unit" : "pixel",
"name" : "x",
"type" : "space"
} ],
sbesson@Sebastien-GS-MacBook-Pro-2025 bioformats2raw % diff -ur mri.zarr mri_pr289.zarr
diff -ur mri.zarr/0/.zattrs mri_pr289.zarr/0/.zattrs
--- mri.zarr/0/.zattrs 2025-09-04 08:45:28
+++ mri_pr289.zarr/0/.zattrs 2025-09-04 08:54:19
@@ -11,15 +11,12 @@
"name" : "c",
"type" : "channel"
}, {
- "unit" : "pixel",
"name" : "z",
"type" : "space"
}, {
- "unit" : "pixel",
"name" : "y",
"type" : "space"
}, {
- "unit" : "pixel",
"name" : "x",
"type" : "space"
} ],The proposed changes are consistent with the discussion in #144 and feels like a reasonable approach given the behavior of other tools in the NGFF ecosystem. /cc @ome/ngff
Could we add a unit test using fake files to capture the expected behavior?
Looks like the team mention might not have worked (maybe due to different organization?) - cc @joshmoore.
Done in 4b39f7e. |
2 participants
Fixes #288.
This lists valid space and time units per https://ngff.openmicroscopy.org/latest/#axes-md, and rejects anything not included in those lists. As indicated in #144,
pixelis one of several units that are supported by OME-XML schema but not NGFF.I debated whether to implement this by defining NGFF schema units, or defining just units that don't overlap between the two schemas (as in #144 (comment)). I think the former approach is going to be easier to keep up to date, but happy to hear other opinions.
With the test data mentioned in #288,
bioformats2raw --debug mri.ome.tiff mri.zarrshould log thatpixelunits are being omitted, andmri.zarr/0/.zattrsshould show nounitattribute for anything underaxes.