How to extract features of an image from a trained model

[ptrblck]

Double post from here. Let’s stick to a single thread to avoid confusion.