Step 4 is: use a pretrained network or train a model from scratch
However in the lesson material we always train a model from scratch, because that way we can explain the concepts that we want to introduce. When I teach I like to show at step 4 in episode 4 how using a pretrained or existing network architecture works by googling for 'cifar10 state of the art keras'. The first hit for me is this keras implementation of a SOTA CNN that is still quite understandable if you are new to keras.
Step 4 is: use a pretrained network or train a model from scratch However in the lesson material we always train a model from scratch, because that way we can explain the concepts that we want to introduce. When I teach I like to show at step 4 in episode 4 how using a pretrained or existing network architecture works by googling for 'cifar10 state of the art keras'. The first hit for me is this keras implementation of a SOTA CNN that is still quite understandable if you are new to keras.
I suggest to add an instructor note about this.