Isaac ROS centerpose performance issues

We have setup the isaac ROS Centerpose pipeline with a pretrained model of book. Tried inference on the realsense camera image topic (used default camera setup)

  • Observations are

    • Detected object pose is not stable, not able to track the pose correctly having lot of jitters.

    • Bounding box generated is at offset position from the actual object.

    • Estimated orientation is not following the actual orientation of an object.

    • In centerpose/detection topic, it is observed that the objects are detected as Ketchup category in both (book as well as bottle) model.

What could be the possible solution to fix these issues. ?

Hello @chiragkumar.makwana1,

Welcome to the Isaac ROS forum! Thank you for sharing your implementation.
To obtain stable results, please keep the object stationary and sample poses for 10–20 seconds. The standard deviation should drop significantly after preprocessing alignment and rectifying intrinsics.
For the CenterPose model, you can reference this page. It is trained with the Objectron dataset, so please ensure you do not include retail-class variants. Also, please follow this guide to address the offset problem. Good luck with your implementation!

Hello,

As there haven’t been any recent updates on this discussion, we’ll be closing the thread to keep the forum organized.

If you continue to experience this issue or have new questions, please start a new topic and include a reference to this discussion for context. This will help our team and other community members assist you more effectively.

Thank you for your participation in the NVIDIA Isaac ROS community.