Support transforms

[jo-mueller]

Hi @aafkegros ,

I just tried MicroscopyNodes for the first time and it's 🤯 . With transformations on the horizon soon, I was wondering whether this is something you would support over here as well?

I haven't browsed the repo here to fully understand where/how the metadata parsing is happening, but would love to contribute if that's on your roadmap :)

[aafkegros]

Yes, this is definitely in the works!

Effectively, I'm waiting for a library like ome-zarr-models-py to easily parse all transforms into affine matrices, which can be applied to the .vdb files i load.

My initial plan was to bake them into the local .vdb file, but i think this will be unflexible (having to duplicate large local files when changing coordinate system) but with the upcoming Blender 5.0 Volume Grid nodes, this can be dynamically applied in the blender scene.

This will then also come with some UI considerations, where i'm imagining hiding the pixel size/axis order panel when loading a zarr with coordinate spaces defined, to replace this with a coordinate space/Scene selector. I'm also imagining only exposing the user to the top-level spaces provided and not discovering through the tree (looking at the future in RFC-8).

The other big UI questions are:

• how to handle the selection of a static max-size of the array when you have differently sampled subarrays. (i'd rather not make an infinite list of dropdowns to click through for loading one file) - maybe making a MAX size loader?
• do all channels of all arrays get listed in the channel selector box? I would assume so, but that might create some weird scenarios. Maybe they should be divided somehow, but that does require some fiddling with the Blender UIList that i'm already stretching to a limit

This comes subsequent to a big Microscopy Nodes 3.0 rewrite i'm doing to take advantage of the Volume nodes in Blender more, which will take still a bit of time.

[aafkegros]

Where long term i think it would also be very interesting to write out the Blender scene back to Zarr (for e.g. manual stitching, annotation etc) - but that will still take a while

[aafkegros]

For contributing, I think this will be very welcome, but might have to wait until Microscopy Nodes 3.0 is out/further along, as some of my core data models will then be formalized better.

In this rework i'm also now starting to bundle nodes in an attached blend file (see #159) where i already have normalization/rescaling to original values of the volumes implemented.

Here a grid transform node could easily fit

With a little bit of work this could actually also be an enum that gets generated upon first load based on the available coordinate spaces, and can be dynamically set on runtime. (this would have to also be mirrored on the labelmask meshes, though)