A bit of a beginner. I was wondering how transfer learning works when the pre-trained model does not contain the classes that you are trying to classify?
For example, lets say the pre-trained model is trained on data that classifies different animals and you are trying to train a model to classify x-ray images.
Should transfer learning even be used in such a case? If yes, why does it make sense?
Should we freeze the layers of the pre-trained model and only train the dense layers? Should we train the whole model?