I'm training a GAN, and I use MSELoss as the GANLoss. But the discriminator's loss of fake and real inputs are totally same after several steps, loss_fake decreased to 0.25, loss_real increased to 0.25 at the same step. Then, they didn't change anymore. So, is my discriminator too strong or too weak?
As for the generator's loss, only SmoothL1Loss changed.