How to generate synthetic pose data from Isaac Sim for objects with symmetries

Hello,

I have been successful in generating synthetic pose data from the Issac Sim 2023.1.1 based Pose Estimation Synthetic Data Generation. Despite this when training the DOPE model with the synthetic data the results are lackluster. From some tests, I believe this is due to the presence of symmetries in my object. Which then complicates the model’s learning of the keypoints, since pixel wise different orientations are equal/highly similar. In the original repository, they provide a data generation method for this problem. I was wondering if the implementation on Issac Sim provides any similar features?

Thanks in advance.

Bumping for visibility

Thank you for posting this, our apologies we couldn’t reply earlier. Let us know if you still need help with this topic. Synthetic data generation is customized using Replicator. When using Isaac Sim, you can import Replicator and summon its annotators.