Unsupervised Projection Networks for Generative Adversarial Networks
This work addresses the challenge of latent space manipulation for generative models, offering a method for downstream applications, but it appears incremental as it builds on existing StyleGAN frameworks.
The authors tackled the problem of projecting images into the latent space of a pre-trained generator, using unsupervised learning to train projection networks, and applied this to StyleGAN for tasks like image super-resolution and semantic clustering.
We propose the use of unsupervised learning to train projection networks that project onto the latent space of an already trained generator. We apply our method to a trained StyleGAN, and use our projection network to perform image super-resolution and clustering of images into semantically identifiable groups.