I am running a model using transfer learning with pytorch in jupyter notebook on jetson tx2. However, the model is not learning at all keeping the accuracy same all the time. When I transferred the same code in google colab, it is working and learning as expected. Please help.
It sounds that the transfer learning source may not support pyTorch model.
May I know which sample do you use first?
I am using densenet121 available in the torchvision models.
densenet121 is a heavy model. TX2 may not have enough computing power for re-training the model.
Please noticed that Jetson platform is designed for fast inferencing.
It’s NOT suitable for training due the memory and bandwidth limitation.