search

yunjey / stargan

StarGAN - Official PyTorch Implementation (CVPR 2018)
MIT License
5.21k stars 967 forks source link

interpolation and extrapolation #32

Open hma02 opened 6 years ago

hma02 commented 6 years ago

Is there a way to interpolate/extrapolate the latent space? In order to see the distribution near the inferenced image y given input image x and condition label c.

Will interpolating on c work? The input is currently (h, w, 3 + nc). With the extra nc channels of all zeros or all ones, I'm thinking if there's a way to use them for interpolation.

satyenrajpal commented 6 years ago

You can maybe look at my code. I've managed to break the generator and interpolate between the original and target latent space in my fork of StarGAN and branch v2 (https://github.com/satyenrajpal/StarGAN/tree/v2). There's a function called 'interpolate' in solver.py. The project is currently in progress but the interpolate function is working.