search

nashory / pggan-pytorch

:fire::fire: PyTorch implementation of "Progressive growing of GANs (PGGAN)" :fire::fire:
MIT License
819 stars 134 forks source link

Some questions about using the code! #14

Open leenoix opened 6 years ago

leenoix commented 6 years ago

Hi there, thanks a lot for providing such wonderful code! I was trying to run your code, but after I run the command for training, which command should I use to generate new images from trained model??

nashory commented 6 years ago

Hi, I newly updated "fake image generation logic using linear interpolation". You can generate interpolated images. Please check 'generate_interpolated.py' file.

I found lots of save/load model bugs while I implement this functionality. Thank you! :)

leenoix commented 6 years ago

Thanks a lot! One last quesiton... Is there any way I can generate interpolation between two 'real' images?

nashory commented 6 years ago

As far as I know, the answer is "no". For us to have interpolated images between two 'real' images, we need to know 'what noise vector(z)' could generate those real images so that we can interpolate. However progressive growing gan doesn't have this inverse inference process (real image → noise).

If you are looking for GAN which can do that, I recommend you to see ALI paper. They have explored this issue.

RahulBhalley commented 6 years ago

Can u please provide the link to ALI paper?

nashory commented 6 years ago

please check this: https://arxiv.org/pdf/1606.00704