Repeat assignments to same scene variable make rendered graphic upside down

This is a weird bug, and I can’t figure out why it is happening. If I want to loop over a few brain areas and create videos from each one, in theory I should be able to do this:

from BrainRender.scene import Scene
from BrainRender.Utils.videomaker import VideoMaker

regions = ['PRT', 'SCm', 'SCs']
for region in regions:
    scene = Scene()
    scene.add_brain_regions([region], colors='white', alpha=0.5)
    vm = VideoMaker(scene, savefile=f"Output/Videos/{region}.mp4", niters=120)
    vm.make_video(azimuth=3)

Indeed, this does create videos of these brain regions. However the second and third (and further videos if there are more regions) are rendered as upside down. I thought perhaps there is an object inside one of the imported files that is being manipulated at some point, but this wouldn’t explain the observation that this script does work:

from BrainRender.scene import Scene
from BrainRender.Utils.videomaker import VideoMaker

scene1 = Scene()
scene1.add_brain_regions(['PRT'], colors='white', alpha=0.5)
vm1 = VideoMaker(scene1, savefile=f"Output/Videos/PRT.mp4", niters=120)
vm1.make_video(azimuth=3)

scene2 = Scene()
scene2.add_brain_regions(['SCm'], colors='white', alpha=0.5)
vm2 = VideoMaker(scene1, savefile=f"Output/Videos/SCm.mp4", niters=120)
vm2.make_video(azimuth=3)

scene3 = Scene()
scene3.add_brain_regions(['SCs'], colors='white', alpha=0.5)
vm3 = VideoMaker(scene1, savefile=f"Output/Videos/SCs.mp4", niters=120)
vm3.make_video(azimuth=3)

Ain’t that weird? Are these two scripts not functionally equivalent? Am I being an idiot and missing something obvious?

[FedeClaudi]

Hi Marco,

I’ve update vtkplotter but the problem is still there. Running this minimal example reproduces the bug:

from brainrender.scene import Scene

regions = ['PRT', 'SCm', 'SCs']
for region in regions:
    scene = Scene()
    scene.add_brain_regions([region], colors='white', alpha=0.5)
    scene.render()

The only way to avoid this is to add to scene.render() this:

if len(vtkplotter.settings.plotter_instances) > 1:
    self._rotate_actors()

to rotate all actors in the scene if there are other plotter_instances.

[FedeClaudi]

Hi,

thanks for looking into this.

As far as I can tell vtkplotter is still inverting the actors for the second and third scene. To correct for this, in brainrender I rotate the actors before rendering the scene. Within the Scene.render() I check if other plotter instances exist, and if they do I rotate the actors (thus correcting the rotation due to vtkplotter:

if len(vtkplotter.settings.plotter_instances) > 1:
    self._rotate_actors()

This is why the output of your test looked okay. Sorry I should’ve explained that better. If you clone the brainrender repo and comment out these lines from Scene.render() you’ll find that the bug Is still there.

As far as I am concerned this solution is okay for brainrender. It might slow rendering for complicated scenes, but I don’t think that it’s an issue that is going to come up frequently. I thought I’d let you know anyways so that you can look into what in vtkplotter is causing this if you want to.

Cheers, Federico