Avirup Dey

Thanks @twentyfiveYang. This post helped. I guess many would keep the same directory structure as in CycleGAN as nothing is mentioned in the official readme.

I faced a similar issue while testing the FFC block separately. ``` fo_conv = FFC(in_channels=3, out_channels=3, kernel_size=3, ratio_gin=0.5, ratio_gout=0.5) x = torch.randn((1,3,256,256)) print(x) print(x.shape) out1, out2 = fo_conv(x) print(out1.shape, out2.shape)...

I'm getting the following error after installing the said cuda toolkit: ``` cannot import name 'PILLOW_VERSION' from 'PIL' ``` Please help.

> And I want to know about preprocessing when evaluating. In the paper, there is face detection and alignment. What is this alignment? It's not only removing the background? The...

I think it's specified in the forward pass of models/ddpm/diffusion.py...after the downsampling steps and applying midblock conv+attention.

@ChanHyeok-Choi "GDrive might still deny you the download, but if you attempt multiple times over the course of several days, probably you'll be able to download it all eventually." :)

Even 128x 128 x 30 is causing OOM on a 24GB A5000. I think without ComfyUI it's impossible to run locally. Maybe a card with >32GB can work