github-pengge

Thx.

So do I. Cannot figure out why.

I didn't notice that. Thanks.

Can't figure out why. In my experiment, when fading in 256x256, the model crashed.

Try `train_no_tanh.py` with latest `data.py` script.

Use `train_no_tanh.py` instead.

Sounds like a good idea, but first, we have to verify whether PG-GAN is helpful in image-to-image network structure.

This is a trick taken from official code.

You can see that old version was commented out. If you need it, just uncomment it.

Changing latent_size is possible, if you also change fmap_max at the same time. Latent_size will be seen as channels at the input block of Generator. So they must match.