Past Recording
Share
Star()
A Style-Based Generator Architecture for Generative Adversarial Networks
Thursday Jan 24 2019 23:30 GMT
Please to join the live chat.
A Style-Based Generator Architecture for Generative Adversarial Networks
Why This Is Interesting

We 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 (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair), 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, we propose two new, automated methods that are applicable to any generator architecture. Finally, we introduce a new, highly varied and high-quality dataset of human faces.

Time of Recording: Thursday Jan 24 2019 23:30 GMT