search

NVIDIA-ISAAC-ROS / isaac_ros_pose_estimation

Deep learned, NVIDIA-accelerated 3D object pose estimation
https://developer.nvidia.com/isaac-ros-gems
Apache License 2.0
165 stars 27 forks source link

Output unstable #11

Open botka1998 opened 1 year ago

botka1998 commented 1 year ago

I trained DOPE for a custom object and have the problem of the estimated poses being jumpy and unstable, also often detecting objects that aren't there.

The original implementation of DOPE had a few parameters that helped stabilize this # Config params for DOPE thresh_angle: 0.5 thresh_map: 0.01 sigma: 3 thresh_points: 0.1

However I can't find anything like this in the ROS2 implementation, could you please provide me with some instructions or information on how to approach this problem?

hemalshahNV commented 1 year ago

These were constants in an older release (here) but are missing as parameters in the latest release. We'll restore these as configurable parameters correctly in our next release. Thanks for pointing this out!

botka1998 commented 1 year ago

Thank you for your help, could you point me to some resources to help me better understand these parameters?