search

RameenAbdal / StyleFlow

StyleFlow: Attribute-conditioned Exploration of StyleGAN-generated Images using Conditional Continuous Normalizing Flows (ACM TOG 2021)
http://rameenabdal.github.io/StyleFlow
2.42k stars 342 forks source link

Thanks for you work, And I want to ask for advises about styleflow used for image-to-image translation. #116

Open vicentowang opened 2 years ago

vicentowang commented 2 years ago

all I needed is image-to-image translation with attribute-editting function, I first use pix2pix，but I cannot edit it for having no latent-space in pix2pix model. then I plan to try pix2style2pixl and styleflow, but there is problem that this two paper both use a pretrained stylegan with a trained encoder, may it possible to train both the encoder and stylegan in an end-to-end fasion ? or there is any other solution or paper for this problem. Thanks a lot.

RameenAbdal commented 2 years ago

Relevant to StyleFlow, the required information for image-to-image can be an input to the conditional part of the Flows. You might have to change the architecture of the neural network (function) inside the flow module to adapt to the spatial information.