search

igul222 / improved_wgan_training

Code for reproducing experiments in "Improved Training of Wasserstein GANs"
MIT License
2.35k stars 670 forks source link

SubpixelConv2D or Deconv2D? #32

Closed LynnHo closed 7 years ago

LynnHo commented 7 years ago

@igul222 Thanks for your good code!

I am confused about why you use SubpixelConv2D for shortcut upsampling but not Deconv2D? SubpixelConv2D brings better results?

igul222 commented 7 years ago

it should work better, but it's also slower. using it only in the shortcuts is a compromise.