f37d0f2da815d3674dbba5ef1d8711357348c94f,implementations/wgan/wgan.py,,,#,88

Before Change

            real_validity = discriminator(real_imgs)
            real_validity.backward(valid)
            // Train on fake images
            fake_validity = discriminator(fake_imgs)
            fake_validity.backward(fake)

            d_loss = real_validity - fake_validity

After Change

            // Generate a batch of images
            fake_imgs = generator(z)
            // Adversarial loss
            loss_D = -torch.mean(discriminator(real_imgs)) + torch.mean(discriminator(fake_imgs))

            loss_D.backward()
            optimizer_D.step()

Instances