Early Stopping for GAN

How can I add early stopping technique for a GAN model ? Should be added for both generator and discriminator ?

The early stopping technique works as a regularizer by stopping updates once some condition is matched (usually no improvement in the validation loss for some epoch).

You can set some conditions for your G or D and stop gradient flow (stop model updating) based on that.

So, it is enough to be applied for just one model (G or D)?

Yes, it is. You have the flexibility to apply it on G or D or both. You should have two optimizers (one for G and another for D).