Is there a ROS2 package for the TLT CV inference pipeline

[Mactemium]

Hello,

I am trying to implement a teleoperation demo using the GestureNet model provided by Nvidia Transfer Learning Toolkit Computer Vision Inference Pipeline ([Overview — Transfer Learning Toolkit 3.0 documentation]) and a UR robot. I was wondering if there was already a ros2 package compatible with the pipeline like it has been the case for the trt_pose_hand model available on Nvidia-AI-IOT gitHub ([GitHub - NVIDIA-AI-IOT/ros2_trt_pose_hand: ROS2 package for trt_pos_hand, "Real-time hand pose estimation and gesture classification using TensorRT"]). If not, could I be able to create this package myself ? I am new to ros2 and don’t know if it would be compatible with the C++ SDK of TLT CV inference pipeline.

Also, I am working on a Jetson Xavier AGX and I was wondering if I would be able to perform transfer learning on the GestureNet model using TLT 3.0. I know the Jetson doesn’t meet the recommended requirements listed here : Requirements and Installation — Transfer Learning Toolkit 3.0 documentation (nvidia.com) but could it work on it even if it is slower to perform ?

Thank you for your time,

Mathis

[dusty_nv]

Hi @Mactemium, you can check the ROS2 packages/containers currently available on this page:

https://nvidia-ai-iot.github.io/ros2_jetson/

If you have further questions about creating new packages, I would recommend to post them as a GitHub Issue to the ros2_jetson GitHub. Thanks!

[dusty_nv]

@Mactemium

Also, I am working on a Jetson Xavier AGX and I was wondering if I would be able to perform transfer learning on the GestureNet model using TLT 3.0. I know the Jetson doesn’t meet the recommended requirements listed here : Requirements and Installation — Transfer Learning Toolkit 3.0 documentation (nvidia.com) but could it work on it even if it is slower to perform ?

Also I don’t believe that TLT supports transfer learning onboard Jetson - TLT runs on an x86 PC with discrete GPU.