NVIDIA researchers came up with a new method to create photo-quality human faces. The chip designer let a new artificial intelligence (AI) algorithm loose on a set of 30,000 celebrity images and the neural network succeeded in creating pretty convincing output with a resolution of 1024 x 1024 pixels. The generative adversarial network (GAN) ran for 20 days on a single NVIDIA Tesla P100 GPU.
A GAN essentially uses two neural networks, which are computing systems modeled on how biological brains actually work. Then it pits them against each other to complete a certain goal.
For the Nvidia project, one neural network was devoted to generating the life-like images. The other network acted as a critic; it flagged which photos were accurate or not. As more images were judged between good and bad, the GAN fine-tuned its approach.
You can see some of the results in the video below.