I am going through the Jetson Inference github and so far so good. Only thing is, the training takes a LONG time. Is there any way to do the training on a PC (with a powerful GPU) and transfer the results to the Jetson?
(I recall that the DIGITS tutorial used to do this but that is now deprecated.)
Thanks.
Hi @jetsonnvidia, yes indeed you can run the PyTorch training scripts on a Linux PC with a GPU. The PyTorch code is organized into git submodules so that you can check-out just that code on your PC if you desire:
Otherwise, you should be able to follow the training guides from Hello AI World just the same. You can export the ONNX model on your PC, and copy that ONNX model (and labels.txt) over to your Jetson.
On your PC, you will want to have installed the NVIDIA driver, PyTorch, and torchvision. On PC, you would install PyTorch the way from PyTorch.org