GANS

Question 1

What is the step-by-step training loop inside a single GAN epoch?

Answer 1

for each batch (e.g., Step 1, Step 2, … Step 500) within the epoch:

Train the Discriminator (D):

FREEZE the Generator (G).

Give D a batch of real images (labeled “real”) and a batch of fake images from G (labeled “fake”).

Update D’s weights to get better at classifying.

Train the Generator (G):

FREEZE the Discriminator (D).

Have G produce a new batch of fake images.

Update G’s weights based on how well it fooled the (now-frozen) D.