Facebook, Inc.

Soumith Chintala

Robert D Fergus

Arthur David Szlam

Emily Denton

In one embodiment, a method includes accessing a plurality of generative adversarial networks (GANs) that are each applied to a particular level k of a Laplacian pyramid. Each GAN may comprise a generative model Gand a discriminative model D. At each level k, the generative model Gmay take as input a noise vector zand may output a generated image {tilde over (h)}. At each level k, the discriminative model Dmay take as input either the generated image {tilde over (h)}or a real image h, and may output a probability that the input was the real image h. The method may further include generating a sample image &#x128;from the generated images {tilde over (h)}, wherein the sample image is based on the probabilities outputted by each of the discriminative models Dand the generated images {tilde over (h)}. The method may further include providing the sample image &#x128;for display.

Growing community of inventors

Jersey City, NJ, United States of America

Soumith Chintala