Poseclassification and action recognition using python code with tensorrt

Accelerated Computing Intelligent Video Analytics TAO Toolkit
, , , , , ,
1

I’m fairly new to Nvidia/ML etc, I use python mostly my goal is to use the deployable model onnx of this

and run inferences on my Jetson orin, I believe I got to use trtexec to get the tensorrt file, can you guys help me with running the tensorrt file(I was able to generate the tensorrt file for another project) using python code, also I would appreciate if you give the trtexec command.

Thank you very much!

• Hardware Jetson Orin

  • TAO - 5.2
  • Deepstream 7
3

Please refer to TRTEXEC with PoseClassificationNet - NVIDIA Docs.

Please refer to PoseClassificationNet - NVIDIA Docs.
The TAO Triton Apps provide an inference sample for Pose Classification.

4

thanks for the trtexec, the triton apps seems like it works as client/server, the server is in cpp, is there a standalone code i could use in python?

5

You can ignore the client/server. Just to leverage how it is doing preprocessing and postprocessing. The history commits can be found in Commits · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub. For example, postprocessing is in tao-toolkit-triton-apps/tao_triton/python/postprocessing/pose_classification_postprocessor.py at b88e9b2af7933f451a03cd35f605f09f54c0aae5 · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub.

You can also leverage the preprocessing and postprocessing about how to run inference against the .pth file.
See tao_pytorch_backend/nvidia_tao_pytorch/cv/pose_classification/scripts/inference.py at main · NVIDIA/tao_pytorch_backend · GitHub.