As I checked NVIDIA github(https://github.com/NVIDIA-Jetson/tf_to_trt_image_classification), I now understand how to implement tensorrt on tensorflow model.
As far as I understand from the repository, I think it can be possible to execute tensorrt engine without installing tensorflow on TX2 if it is done that exporting to intermediate file such as .plan or .uff. At this point, I am wondering if it is right.
please let me know if it is right.
Correct.
With UFF or PLAN(built on TX2 architecture) file, you can launch TensorRT engine without TensorFlow installed.
Thanks.