Cannot use TAO Deploy in Jetson AGX Orin

[Morganh]

@bobchen1017

Traceback (most recent call last):
  File "scripts/gen_trt_engine.py", line 24, in <module>
    from nvidia_tao_deploy.cv.grounding_dino.engine_builder import GDINODetEngineBuilder
ModuleNotFoundError: No module named 'nvidia_tao_deploy.cv.grounding_dino'

Because it is not found in your working path or your PYTHONPATH by export PYTHONPATH=$PYTHONPATH:/app/tao/tao_deploy/nvidia_tao_deploy.
You can check nvidia_tao_deploy.cv.grounding_dino.engine_builder and confirm it is available.
Also, if needed, please note that you can modify the existing 5.5 code to make it work in 5.0 deploy docker.

BTW, for DINO inference in Jetson device, one existing way is to use deepstream_tao_apps github. The PeopleNet Transformer is actually trained by DINO. Config file can be found in
deepstream_tao_apps/configs/nvinfer/peoplenet_transformer_tao/pgie_peoplenet_transformer_tao_config.txt at master · NVIDIA-AI-IOT/deepstream_tao_apps · GitHub.
Doc is in Deploying to DeepStream for DINO - NVIDIA Docs.