Input and output channels to transpose convolution

Course Q&A Generative Adversarial Networks (GANS) Build Basic Generative Adversarial Networks
1

Why are we changing the input and output channels as we are going to the next layer in the Discriminator and Generator?

Thanks,
Sailesh

3

Hey @saileshbaidya,

Usually, the idea behind it is to help both the Generator and Discriminator to succeed their task. In the Generator, this helps building complexity and make it more likely to generate complex data, on the other hand, in the Discriminator, this helps reducing complexity so the task of deciding if it is real or fake gets easier.

I hope this answers your question.

Cheers,
Lucas

4

Thanks @lucas.coutinho. I get the general idea. But, still not clear why the output channel increases for all the hidden layers in the discriminator while we are decreasing them in the generator’s hidden layers.

Thanks,
Sailesh