A style-Based Generator Architecture for Generative Adversarial Networks
The authors propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature. The new architecture leads to an automatically learned, unsupervised separation of high-level attributes, and it enables intuitive, scale-specific control of the synthesis. The new generator improves the state-of-the-art in terms of traditional distribution quality metrics, leads to demonstrably better interpolation properties, and also better disentangles the latent factors of variation. To quantify interpolation quality and disentanglement, the authors propose two new, automated methods that are applicable to any generator architecture. Finally, the authors introduce a new, highly varied and high-quality dataset of human faces.