Using Python3.7 instead python 3.6 on jetson TX1 and transfer installed Libraries

Autonomous Machines Jetson & Embedded Systems Jetson TX1
1

I installed JetPack 4.6.3 on Jetson TX1 with Nvidia SDK manager and the SDK installed CUDA 10.2, tensorrt 8.2, and cuDNN 8.2 on the board.
but no version of tensorflow is compatible with Python 3.6 and cuda 10.2 !!!
what can I do now?
I want to convert the tensorflow SSD-mobile net to a tensorrt model for inferencing, but nothing has been installed on the board ?!
the uff and tensorrt libraries are installed on Python3.6 and I can’t transfer them to the new python because of conflicts…

3

Hi,

You will need an ARM-based package for TX1, this can be found in our page below:
https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html

If you just want to deploy the model with TensorRT, a recommended way is to convert it into ONNX format on an x86 environment.
Then copy the file to Jetson, which can be deployed with the TensorRT binary directly.

$ /usr/src/tensorrt/bin/trtexec --onnx=[file]

Thanks.

4

1-Is this command for inferencing or for converting to TensorRT?
I have to draw bounding boxes…
2-Do I need to install a new version of Python (Python 3.7)?
Python 3.6 is not ok?

5

I would be grateful if you could guide me

6

@AastaLLL
I have converted the model to .onnx format and also convert the .onnx model to TrtEngine
trtexec command gives me a .trt file,
now
what can I do ?!
how can I infer the .trt file ?! (loading the .trt file, making predictions, and drawing bounding boxes)
the command I try:
/usr/src/tensorrt/bin/trtexec --onnx=./home/deep/Project/project/Dry.onnx --saveEngine=/home/deep/Project/ --fp16
this script gives me a .trt file…

7

I would be grateful if you could guide me

8

Hi,

The trtexec binary will convert the model and run inference for benchmarking.
If you want a python based sample, you can check below:
https://elinux.org/Jetson/L4T/TRT_Customized_Example#TensorRT_Python

We only provide the python 3.6 packages for Ubuntu 18.04.
For other versions, you can build the pybinding from the source:

Thanks.

10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.