Nvidia has released a mesmerising and creepy video of a neural network that renders photo-realistic faces.
Coinciding with the video, the tech company also released a paper on Friday which explains the new method of generating unique faces via a generative adversarial network (GAN).
It is best described as a class of algorithm where researchers pair two competing neural networks against each other.
According to Tom Mckay of Gizmodo, “one of the two neural networks within the GAN is put to a generative function (such as rendering images or trying to solve a problem) whilst the other network is put in an adversarial role, challenging the first’s results.
“The intent is that the generative neural network will produce a superior result by bouncing its ideas off of the adversarial counterpart.”
Nvidia shared with their Twitter followers just how computer scientist Ian Goodfellow created GAN.
— NVIDIA AI (@NvidiaAI) June 7, 2017
The Nvidia team also wrote that with a new progressive training method, they were able to generate “images of unprecedented quality” using the CelebA-HQ database of photos of famous individuals. So the celebrities that are normally familiar to us are completely unrecognisable in this video as they transition from one to another.
Nvidia’s method has also allowed for great generation of objects and scenery, which you can see for yourself in the following video:
McKay added: “Back in 2015, Google released DeepDream, another neural network that users put to work rendering anything from disturbing dog collages to even more disturbing psychedelic porn.
“DeepDream may have just been the prelude to less aesthetically off-putting but much more significant applications of the slightly creepy technology – such as generating photo-realistic, high-definition images of people who never existed.”