search

akanimax / BMSG-GAN

[MSG-GAN] Any body can GAN! Highly stable and robust architecture. Requires little to no hyperparameter tuning. Pytorch Implementation
MIT License
629 stars 105 forks source link

MSG-cycleGAN #47

Open LoveU3tHousand2 opened 2 years ago

LoveU3tHousand2 commented 2 years ago

I try to implement msg way to cycleGAN for two weeks. The generator : Unet-withSkipConnection and downsample(k=4, s=2 conv) upsample(Deconv) 8x.

But the D_loss, both D_A and D_B always decreased near to zero and G_loss(lsgan) always stuck in 0.01 ...

What I want to know is whether this combination will work? msg + cycleGAN? Anybody tried before?

thuangb commented 2 years ago

As answered in #40, the D loss should be near zero. What matter most in training GAN is not the loss but the visual quality, if you see it is good then just keep training

LoveU3tHousand2 commented 2 years ago

Thanks for your replying, but msg strategy and cyclegan just can not work together...I already gave it up..

Sent from my iPhone

On Jun 17, 2022, at 12:19, thuangb @.***> wrote:



As answered in #40https://github.com/akanimax/BMSG-GAN/issues/40, the D loss should be near zero. What matter most in training GAN is not the loss but the visual quality, if you see it is good then just keep training

— Reply to this email directly, view it on GitHubhttps://github.com/akanimax/BMSG-GAN/issues/47#issuecomment-1158474080, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AQTQZ5BGSKRROOC433P5KHTVPP4DTANCNFSM5Q6YVFYA. You are receiving this because you authored the thread.Message ID: @.***>

keruoya commented 8 months ago

hi ,I also want to try to do it , I want to ask that is this true that they can't work together?