ISF-GAN: An Implicit Style Function for High Resolution Image-to-Image Translation

Recently, there has been an increasing interest in image editing methods that employ pre-trained unconditional generators (e.g., StyleGAN). However, applying these to translate images multiple visual domains remains challenging. Existing works do not often preserve the domain-invariant part of identity human face translations), or they usually handle allow for multi-modal translations. This work proposes implicit style function (ISF) straightforwardly achieve and multi-domain image-to-image translation from generators. The ISF manipulates semantics a latent code ensure generated manipulated lies desired domain. Our faces animal manipulations show significantly improved results over baselines. model enables cost-effective unsupervised translations at high resolution using GANs.