Utilising learning rate discrimination & progressive resizing