something is wrong when computing 3D bounding boxes for certain objects

Omniverse Isaac Sim
1

Hi, I was creating a dataset with the ground truth of object’s 3D bounding box. I observed that certain objects did not get correct 3D bounding boxes or even were totally missing. Please have a look at the attached image. Hope this issue can be fixed in the future version.

IsaacSim_bug.png
IsaacSim_bug.png903×1003 942 KB

2

Hi xmaster,

THanks for pointing out, I will pass this to the development team.

Kindly,
Liila

3

Hello,

is there any progress on this issue?

I have the same problem ( Possible Bug: BoundingBox for prims with geometries with skeleton animation are to large - Omniverse / Developer - NVIDIA Developer Forums) and that is a real showstopper for us, because we rely on bounding box computation for relative positioning objects to each other…

Thanks for your help

Carl

4

Hi @ahaidu - Can you please look into this?

5

Hi there, can you elaborate what is the issue you are encountering, since the initial question is an old one (from 2019) and might be unrelated.

If it has to do with 3d bounding boxes from annotators, here is some information on how to use or visualize them live:

If you need bounding boxes at the USD level, here is an example using OBB and AABB:

7

Thanks for your response,

I think my problem is some kind of bug in USD… in the meantime I had a deeper look into it, and have created a very small scene to reproduce the problem under my original post ( Possible Bug: BoundingBox for prims with geometries with skeleton animation are to large - Omniverse / Developer - NVIDIA Developer Forums)

It looks like USD does not calculate the bounds correct for meshes with skeleton animations in some scenarios…