I want to train a model with CNN in jetson nano

Hello,

I want to get some metrics by running a CNN on jetson nano. Has anyone run a Convolution Neural Network from a github or any other source that is reliable and runs for over an hour, so that I can get some metrics about time etc? I have tried some but they do not run in embedded jetson nano + Ubuntu OS etc. That is I ask for your help… Thank you!

Hi plaketas1, check out this repo: https://github.com/dusty-nv/jetson-inference

You could run camera inferencing on one of the included pre-trained models for a long time.

Alternatively, if you are interested in training, it includes transfer learning with PyTorch here:
https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-transfer-learning.md

In this link:

https://github.com/dusty-nv/jetson-inference/blob/master/docs/pytorch-plants.md

it says:

We’ll use the same training script that we did from the previous example, located under python/training/classification/. By default it’s set to train a ResNet-18 model, but you can change that with the --arch flag.

However the link: python/training/classification/

does not work. How do I do the training?

I cannot find the

train.py

on folder:

/Desktop/jetson-inference-master/python/training/classification

What is wrong??

Do you see anything under that folder, or is it empty?

If it’s empty, the submodules may not have been cloned - try running this:

$ cd /Desktop/jetson-inference-master
$ git submodule update --init

If all else fails, the code for PyTorch classification training is in this submodule: https://github.com/dusty-nv/pytorch-classification/tree/6b8fcd38fee76cae26e43b9bd547491813bf423d