Chen Hou

Want it too, could anyone share the discriminator weights?

use vae.encode(x).latent_dist.sample() might just work

Yeah, you're right, it's a flip over diagonal, but I think it can be seen as a kind of 90 degrees rotation.

you can set --config to custom, and rewrite the custom config and a dataloader for it in data_utils.py then you can train on single image instead of pre-defined dataset