How to update classification layers without changing weights for convolutional layers

Question

I have a CNN with numerous convolutional layers. To each of these convolutional layers I have attached a classifier, to check outputs of intermediary layers. After losses have b