search

chaiyujin / glow-pytorch

pytorch implementation of openai paper "Glow: Generative Flow with Invertible 1×1 Convolutions"
MIT License
505 stars 79 forks source link

Dataset Issue #10

Closed axium closed 5 years ago

axium commented 5 years ago

Thanks for providing a well document implementation of Glow. I am not able to figure out how did you prepare training data for CelebA. I am using your module CelebADataset but failed to produce similar images as you have shown. I tried this module on both aligned cropped version of CelebA and in-the-wild version as well.

Can you please share you exact data preparation methodology?

chaiyujin commented 5 years ago

Hi, @axium I never did any pre-processing.

axium commented 5 years ago

My bad! What I mean by pre-processing is the alignment of faces, center cropping etc. Could you tell me which version of CelebA dataset did you use? I am interested in those exact images on which you trained your model on. If you help me understand I would grateful.

chaiyujin commented 5 years ago

I use image in dir img_align_celeba. CenterCrop(160) and resize(32)

axium commented 5 years ago

Thanks. Really appreciate your help in this. I was making a mistake. Thanks.