I have a query on how to use intermediate feature map as input to the MLP network while implementation.

In other words Input shape=(32,32,3) is the standard way of input.

How can make use intermediate layers as input to the neural network , in case the inputs from intermediate layers is (None, 32768) , (None, None, None, 2048)

I would like to provide my feature maps as input to discrminator network.