How about RMSprop? - Githubissues

yunjey / stargan

StarGAN - Official PyTorch Implementation (CVPR 2018)

MIT License

5.23k stars 972 forks source link

How about RMSprop? #96

Open miqbal23 opened 5 years ago

miqbal23 commented 5 years ago

Hello, I want to ask about the reason the network optimizers, and since I haven't seen anyone asking for it.

I noticed from your paper (and also implementation) that you use Adam as optimizer for the network. Have you tried using RMSprop instead, since WGAN (which training method you implemented in StarGAN) used RMSprop in their implementation?

yunjey commented 5 years ago

@nobodykid I have never tried to use the RMSprop optimizer. I think it behaves similarly to the Adam optimizer when beta1 is set to 0. CycleGAN also used the Adam optimizer.