Can I change googlenet's algorithm

Can I change googlenet’s algorithm? I say algorithm like https://github.com/TommyHuang821/Deep-Learning-API-example/blob/master/TensorFlow/TF%20範例%20I%20(DNN%20Classification).md.

Hi,

Would you mind to share more information about your requirement?

TensorFlow is supported on the Jetson Nano.
You can install it based on this document and create a network as the link you shared.

Thanks.

Sorry , I shouldn’t share this website:https://github.com/TommyHuang821/Deep-Learning-API-example/blob/master/TensorFlow/TF%20範例%20I%20(DNN%20Classification).md. I want to ask a question .Can I change GoogleNet-12’s loss ,sgd rate?Can I change loss/cost function , network’s weight and bias?

Hi,

You can do this since Jetson do support TensorFlow frameworks.
These change are native support in TensorFlow so surely you can run it on Jetson.

One problem is that loss is more related to a training stage rather than inference.
Since Jetson is an edge device, we don’t recommend to apply a training task on Jetson platform.

Thanks.

My googlenet’s runs on jetson inference 's tensorrt. Can I change GoogleNet-12’s loss ,sgd rate?Can I change loss/cost function , network’s weight and bias?

Hi @andy8902, the Googlenet12 model was trained in PyTorch on ResNet18, you can see the training code here: https://github.com/dusty-nv/pytorch-classification

It is found in your jetson-inference repo under jetson-inference/python/training/classification

So you can edit the PyTorch train.py script to suit your needs.