[BUG] Scale transformation mismatch between image and label

[tcompa]

Describe the bug

Hi there, I found what looks like a small bug in the scale at which a label array is displayed (relative to the corresponding OME-Zarr image). The same dataset loads fine in vizarr.

To Reproduce

1. Run pnpm i & pnpm run dev for current main - 1178401.
2. Open http://localhost:5173/?source=https://raw.githubusercontent.com/tcompa/hosting-ome-zarr-on-github/refs/heads/main/with_labels/20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr/B/03/0 (this is a typical output coming from Fractal, also available at https://zenodo.org/records/10257149 - cc @jluethi).
3. Enable the labels layer.
4. Observe a scale mismatch - see screenshot below.
5. Compare with how vizarr displays the same image+labels at https://hms-dbmi.github.io/vizarr/?source=https://raw.githubusercontent.com/tcompa/hosting-ome-zarr-on-github/refs/heads/main/with_labels/20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr/B/03/0 - see screenshot below.

Note that the image and label multiscales have different YX scale transformations for their highest-resolution level:

• Image: https://github.com/tcompa/hosting-ome-zarr-on-github/blob/af7beee54655b92e6a49d1177bcade1274a99a13/with_labels/20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr/B/03/0/.zattrs#L30-L33
• Label: https://github.com/tcompa/hosting-ome-zarr-on-github/blob/af7beee54655b92e6a49d1177bcade1274a99a13/with_labels/20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr/B/03/0/labels/nuclei/.zattrs#L32-L34

Expected behavior

The same as vizarr.

Screenshots (if applicable)

biongff-viewer:

vizarr:

Environment (please complete the following information):

• OS: Ubuntu24
• Python/Node/Django/etc. version: pnpm 10.17.1
• Browser (if applicable): Firefox

[jluethi]

A comment on what is likely to go wrong here: We often have different scales in our label images compared to the intensity images. Vizarr loads this scale from the multiscale metadata of each and then scales the two accordingly before overlaying them.
If the label image isn't scaled, the screenshot from above is expected

[AlexSurtees]

Hi @tcompa, thanks for the detailed report. The biongff-viewer repository is being deprecated and the work there is being migrated over to the vizarr repository here: https://github.com/BioNGFF/vizarr.

There is a pull request in review for the bulk of this migration here: BioNGFF/vizarr#28, and you can test that pull request here: https://deploy-preview-28--biongff-vizarr.netlify.app/.

I had a go at reproducing the issue you describe using this version and can't seem to reproduce it, so it looks as though it has been fixed since you made this issue. Would you mind double-checking that the issue is resolved using the link above to test?

Cheers,
Alex

[tcompa]

Hi Alex,

I confirm that the issue is not present in the version deployed at https://deploy-preview-28--biongff-vizarr.netlify.app/.

That is, https://deploy-preview-28--biongff-vizarr.netlify.app/?source=https://raw.githubusercontent.com/tcompa/hosting-ome-zarr-on-github/refs/heads/main/with_labels/20200812-CardiomyocyteDifferentiation14-Cycle1_mip.zarr/B/03/0 looks like this:

I guess I can now close this one - thanks!