Is it possible to get long-running animations and textured meshes working in a single environment?

[gth828r]

Hello,

I am trying to model a satellite orbiting a model of Earth. I would like to texture my Earth model, and I would like to view the animation of the satellite orbiting. For now my Earth texture is static, but eventually I would like to periodically change it as part of the animation to reflect changes in weather patterns.

I found 2 different high-level approaches:

1. Use the draw method or O3DVisualizer class and pass in a callback for on_animation_frame (or similar draw and callback functions) where I can apply transformations to objects in the loop
2. Implement a loop where I can apply transformations and render using visualization.Visualizer

In my experience, approach 2 does not have a path to loading textures that works, so I stopped exploring it. Am I missing something there?

Approach 1 works, but I have found it to be unstable when the animation occurs. It always eventually segfaults in the draw method, although the actual segfault doesn't occur in my callback (although I acknowledge that perhaps something I do in the callback causes the issue downstream).

The high level implementation of my callback is as follows:

1. Fetch the existing materials from all existing meshes in the scene for future reference
2. Apply transformations to existing meshes (my callback is a closure, so my originally-defined meshes are in scope)
3. Call scene.clear_geometry()
4. For each mesh in the scene, call scene.add_geometry() passing in the transformed meshes and their textures (obtained in step 1)

I never explicitly delete any meshes or materials in my code, I just remove them from the scene. I have other meshes in my environment to visually track other components (e.g. coordinates, locations of interest on Earth, etc), but when I disable them it takes longer for the segfault to occur. This makes it feel like it could be some resource-usage issue related to calling scene.add_geometry more frequently; however, when I watch memory usage through top and nvidia-smi, I do not see it growing as the animation runs in either case.

Is there something obvious I am doing wrong here? Is there a better way? Is this just not what Open3D was intended for in terms of usage?

Note that I am using Open3d 0.19.0, so this shouldn't be caused by the well-documented numpy 2.0 compatibility issue.