How do I add a Normalize(2) layer (or any other type of layer) after the last (fully connected) layer of Pytorch’s inceptionv3 architecture?
Take a look at this link. It goes about how to add custom layers to a pre-trained network and then further tune them for your specific use case. In order to add new layers, take a look at the section where they explain how to modify the VGG or AlexNet architecture. I’m sure it can be extrapolated for your use case.