Issac ROS for dnn stereo disparity - disparity map accuracy improvement

Greetings to All ,
Currently I am using Orin Nano with JP 5.1.1. I have followed the Quickstart instructions mentioned in the git link to generate the disparity map for stereo cameras.

We were able to generate the output but there were differences between the expected output and the output generated by us. We have used the sample images which was in the source.


The top image (change_ess_output.png) is generated by us and the bottom image (output_raw.png) is the expected output mentioned in the git link.

Why does the output vary ? is there a way to improve the accuracy ?

Sample Images filepath in github = NVIDIA-ISAAC-ROS/isaac_ros_dnn_stereo_disparity/resources/examples/

Git link for isaac_ros_dnn_stereo_disparity :- GitHub - NVIDIA-ISAAC-ROS/isaac_ros_dnn_stereo_depth: Hardware-accelerated, deep learned stereo disparity estimation

Hi @gautham.krishnan

Sorry for this late reply.

Please try our new Isaac ROS 2.1. We made many updates, new documentation, and new accuracy

Documentation: NVIDIA Isaac ROS — isaac_ros_docs documentation

I suggest updating Jetpack, cleaning the environment, and retrying the stereo depth node.

https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_dnn_stereo_depth/isaac_ros_ess/index.html

Best,
Raffaello

Ok , will let you know after following the above steps.
Thank you