Jetson TX2 set up

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Hi,

i am struggling to set up tensorrt on jetson tx2.

i can train model on DIGITS and run it using imagenet-camera provided in jetson-inference .

  • How to install tensorRT on jetson tx2 .(tried to install using debian package but getting errors )
  • which version of ubuntu is required( i am using 16.04 for jetpack 3.3)
    -cuda version required for jetson ( i am using cuda 9.0)
  • which version of tensorflow and opencv ?
2

Hello,

NVIDIA JetPack SDK is a comprehensive solution for building AI applications. TensorRT 4.0 is included with JetPack.

Please reference https://developer.nvidia.com/embedded/jetpack-notes for release details.

3

but while doing import tensorrt , it shows module not found ,am i doing something wrong

import tensorrt
Traceback (most recent call last):
File “”, line 1, in
ImportError: No module named tensorrt

4

Hello, the entire SDK package is flashed to Jetson by default (unless explicitly selected JetPack not to install TensorRT). However the library is called libnvinfer and the header is NvInfer.h.

Here’s a link to a code example using it.

5

thanks a lot i can run the sample classification model from pretrained model from tensorflow github , but any sample code example for running or test model trained by DIGITS tensorflow (as it does not have .pb file ) , i am bit confused as this trained model has to be converted to frozen graph and then trt_graph .