[BUG] Spinal cord mesh rotated at 90 degrees in meshio/napari

[alessandrofelder]

Describe the bug The allen_cord_20um mesh behaves differently to the other meshes. I don't think it's because it's anisotropic. Could also be a meshio (or a napari) bug (but it doesn't happen with other atlases AFAIT)?

To Reproduce

• make a copy of 250.obj and annotations.tiff in a local folder
• rename the copies to mesh_0.obj and image_0.tiff
• copy the python script below to the same local folder.

from pathlib import Path
import napari
import meshio
from tifffile import imread

viewer = napari.Viewer()

scales = [
    [10,10,20], # image 0 has anisotropic scale
]
for i in range(1):
    image_path = Path(__file__).parent / f"image_{i}.tiff"
    image = imread(image_path)
    image = viewer.add_image(image, scale=scales[i], colormap='hsv')

    mesh_path = Path(__file__).parent / f"mesh_{i}.obj"
    mesh = meshio.read(mesh_path)
    points = mesh.points
    cells = mesh.cells[0].data
    surface = viewer.add_surface((points, cells), opacity=0.4)
    
    print(f"image {i}:")
    print(image.extent.world)
    print(surface.extent.world)

if __name__ == "__main__":
    napari.run()

• run the python script in a conda env with napari installed
• switch napari to 3d view
• the mesh and the image are at 90 degrees to each other

Expected behaviour The mesh and the image should align with each other.

Log file

Screenshots If applicable, add screenshots to help explain your problem.

But in paraview it looks reasonable (note the transform in the screenshot is the same as the scale in the example script here) Computer used (please complete the following information):

• OS: macOS Ventura 13.4.1 (22F82) (Apple M2 chip), 16GB RAM

Additional context I expect running the script above on any other annotation+root mesh combo works fine (I've tested only on allen_mouse_100um though), because the script is a simplification of what I added in https://github.com/brainglobe/brainglobe-napari/pull/31 and that is extensively tested on various (but not all) atlases. Maybe the .obj has some metadata that meshio ignores that the other objs don't, and paraview knows how to deal with that? Maybe the spinal cord's obj is somehow special