If one wants to use a pre-trained model on a new dataset, such that the images of the newly used dataset have different dimensions than the ones used with the pre-trained model, how make_model() will prune/adjust the newly defined network. More, how the original classifier trained on ImageNet can be used with Cifar10 via train_original_classifier?
Plus, what is the difference between train_original_classifier and pre_trained_model in make_model()?