A brand new web site referred to as This Individual Does Not Exist went viral this week, and it has one easy operate: displaying a portrait of a random individual every time the web page is refreshed. The web site is pointless at first look, however there is a secret behind its seemingly limitless stream of pictures. Based on a Fb put up detailing the web site, the photographs are generated utilizing a generative adversarial networks (GANs) algorithm.

In December, NVIDIA printed analysis detailing the usage of style-based GANs (StyleGAN) to generate very practical portraits of people that do not exist. The identical expertise is powering This Individual Does Not Exist, which was created by Uber software program engineer Phillip Wang to ‘elevate some public consciousness for this expertise.’

In his Fb put up, Wang stated:

Faces are most salient to our cognition, so I’ve determined to place that particular pretrained mannequin up. Their analysis group have additionally included pretrained fashions for cats, automobiles, and bedrooms of their repository that you may instantly use.

Every time you refresh the positioning, the community will generate a brand new facial picture from scratch from a 512 dimensional vector.

Generative adversarial networks have been first launched in 2014 as a option to generate pictures from datasets, however the ensuing content material was lower than practical. The expertise has improved drastically in only some years, with main breakthroughs in 2017 and once more final yr with NVIDIA’s introduction of StyleGAN.

This Individual Does Not Exist underscores the expertise’s rising capability to supply life-like pictures that, in lots of instances, are indistinguishable from portraits of actual individuals.

As described by NVIDIA final yr, StyleGAN can be utilized to generate extra than simply portraits. Within the video above, the researchers reveal the expertise getting used to generate pictures of rooms and autos, and to change ‘positive kinds’ in pictures, reminiscent of the colour of objects. Outcomes have been, typically, indistinguishable from pictures of actual settings.


(function(d, s, id) {
var js, fjs = d.getElementsByTagName(s)[0];
if (d.getElementById(id)) return;
js = d.createElement(s); js.id = id;
js.src = “http://connect.facebook.net/en_US/sdk.js#xfbml=1&version=v2.7&appId=190565384410239”;
fjs.parentNode.insertBefore(js, fjs);
}(document, ‘script’, ‘facebook-jssdk’));

// Twitter

// G+


Shop Amazon