Different loss than the original paper about dcgan-tensorflow HOT 2 CLOSED

Hi gaohuazuo, from the paper here: https://arxiv.org/pdf/1406.2661v1.pdf , right below equation (1) which defines the loss function, they state

"Rather than training G to minimize log(1-D(G(z))) we can train G to maximize log(D(G(z))). This objective function results in the same fixed point of the dynamics of G and D but provides much stronger gradients early in learning"

If you want to minimize equation (1) directly, you can just write

self.g_loss = self.d_loss_fake*(-1.0)

instead of

self.g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(self.D_logits_, tf.ones_like(self.D_)))

In my experience there's not a big difference.

from dcgan-tensorflow.

Thanks for the discussion and yes I think the equation is same.

from dcgan-tensorflow.