There is some errors in calculating losses, Inconsistent with the papers.
TriLoo opened this issue · 4 comments
For example, formula (9) calculates the loss of two generators where addition is used:
however, I found the corresponding implementation from TF.py
is as following:
netG_train_loss = tf.reduce_mean(netD_netG_train_output1_1) - tf.reduce_mean(netD_netG_train_output2_2)
and I think the code should be:
netG_train_loss = tf.reduce_mean(netD_netG_train_output1_1) + tf.reduce_mean(netD_netG_train_output2_2)
same thing happens to the calculation of Discriminator:
Paper:
code:
netD_train_loss = (-tf.reduce_mean(netD_train_output1_1) + tf.reduce_mean(netD_train_output2_1)) + (-tf.reduce_mean(netD_train_output1_2) + tf.reduce_mean(netD_train_output2_2))
and the correct code may be
netD_train_loss = (-tf.reduce_mean(netD_train_output1_1) + tf.reduce_mean(netD_train_output2_1)) + (-tf.reduce_mean(netD_train_output2_2) +
tf.reduce_mean(netD_train_output1_2))
any one can give me some advices ?
Hi @TriLoo , could you share the code and model linked by @nothinglo in README.md?
He doesn't look like to maintain link of the code and model.
From your issues, I think you probably have the code. Could you share them?
Thank you.
Hi @TriLoo , could you share the code and model linked by @nothinglo in README.md? He doesn't look like to maintain link of the code and model. From your issues, I think you probably have the code. Could you share them?
Thank you.
I have tried to find the code but failed. It has been a long time gone
Hi @TriLoo , could you share the code and model linked by @nothinglo in README.md? He doesn't look like to maintain link of the code and model. From your issues, I think you probably have the code. Could you share them?
Thank you.I have tried to find the code but failed. It has been a long time gone
Thank you for your effort!
I'll have to re-implement it :)