Updating the generator using the joint loss gradient?

[1420108]

Hi, thank you very much for your work. In the paper, they use the joint loss (L2 and GAN loss) gradient to update the generator G, but in your implementation, I notice that you only use L2 loss gradient to update G. Have I missed something here? Thank you.

[tadax]

Thank you for your comment! You are right. I will update and re-train the model in a few days.

[avisekiit]

HI @tadax : can I now use the repo, if you have corrected how the G network is updated.

[roxanneluo]

Hi @tadax, thank you a lot for sharing your code. Do you have any updates on this issue?

[RobinCSIRO]

I think it is easy to add the discriminator loss for training generator:
def calc_g_loss(self, x, completion):
loss1 = tf.nn.l2_loss(x - completion)
loss2 = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=fake, labels=tf.ones_like(real)))
loss = loss1+loss2
return tf.reduce_mean(loss)

[fabrahman]

Hi @RobinCSIRO have you tried that? did you get reasonable results?
I guess he skiped one step in the algorithm. But then again for pretrainig step, we dont need the GAN loss to be incorporated.
Do we have to have two g_loss? one only the l2 norm an the other with l2+GANS_loss