Hi! We are trying to train our dataset on ssd mobilenet v2 model on jetson nano like dusty did on youtube. Aproxiametly each epoch takes 3-4 minutes. We will train the model with like 1000 epoch, it will be very long time. So how can i train on my windows pc and then integrate the model with jetson nano?
Hi @ahmet.cankarakus, you can attempt to install PyTorch for Windows from https://pytorch.org/, however I have not tested this with the pytorch-ssd code. There are other dependencies such as Pandas that you would need to install for Windows as well.
I have used pytorch-ssd on an Ubuntu PC with NVIDIA GPU, and that worked well after installing CUDA/cuDNN/PyTorch and the dependencies in Ubuntu.
I think you may find diminishing returns for training it for that many epochs, typically you only need 100 epochs (or maybe 200) to get full convergence of the model.
1 Like
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.