How to train the Discriminators in the article? #80

New Issue

2021-08-03T04:38:20+02:00

linyu0219 commented

2021-08-03 04:38:20 +02:00

(Migrated from github.com)

I want use your GAN Discriminators and hinge loss in article for my own work?
but the loss can't reduce.
I note you write using adam in article ,the 𝛽1=0 and 𝛽2=0.999, is this has any writting error?
Do you have any tricks training the Discriminators, multi stage？
Stage 1, only hinge loss
Stage 2, hinge loss +10*GP loss?

I want use your GAN Discriminators and hinge loss in article for my own work? but the loss can't reduce. I note you write using adam in article ,the 𝛽1=0 and 𝛽2=0.999, is this has any writting error? Do you have any tricks training the Discriminators, multi stage？ Stage 1, only hinge loss Stage 2, hinge loss +10*GP loss?

NNNNAI commented

2021-08-06 15:05:49 +02:00

(Migrated from github.com)

First, there is no writting error for the beta value.Second, you can try hinge loss +1e-5*GP loss.

ygtxr1997 commented

2022-10-09 10:31:43 +02:00

(Migrated from github.com)

In your code, beta2 is 0.99 which is inconsistent with 0.999 wrote in your paper.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: CalvinBackup/neuralchen-SimSwap#80