search

lucidrains / lightweight-gan

Implementation of 'lightweight' GAN, proposed in ICLR 2021, in Pytorch. High resolution image generations that can be trained within a day or two
MIT License
1.63k stars 222 forks source link

FFHQ results are not matching as shown in the paper #84

Open VirajBagal opened 3 years ago

VirajBagal commented 3 years ago

I trained on 1000 FFHQ 1024x1024 images for a little more than 10 hours with all default settings. Following are some of the samples after training of the network:

9

These images look much worse than those shown in the paper. Anyone knows the reason? Help much appreciated.

I think the default model in the code is same in size as described in the paper

yangyu615 commented 3 years ago

I trained on 1000 FFHQ 1024x1024 images for a little more than 10 hours with all default settings. Following are some of the samples after training of the network:

9

These images look much worse than those shown in the paper. Anyone knows the reason? Help much appreciated.

I think the default model in the code is same in size as described in the paper Can you share FFHQ datasets?Thank you very much!

yangyu615 commented 3 years ago

Can you share FFHQ datasets?Thank you very much!