Ai-Powered Website Generates Realistic Human Faces On the Spot
Written by Sergiu Gatlan / Courtesy of Bleeping Computer
A website created by Philip Wang, an Uber software engineer, and hosted at thispersondoesnotexist.com allows its visitors to generate realistic looking human faces of people that do NOT actually exist each time they hit the Refresh button.
The websites uses artificial intelligence and machine learning and pretrained models for StyleGAN, the official TensorFlow implementation of NVIDIA Research Projects’ “A Style-Based Generator Architecture for Generative Adversarial Networks” research paper.
As Wang said in a Facebook post:
Recently a talented group of researchers at Nvidia released the current state of the art generative adversarial network, StyleGAN, over at https://github.com/NVlabs/stylegan.
I have decided to dig into my own pockets and raise some public awareness for this technology.
Faces are most salient to our cognition, so I’ve decided to put that specific pretrained model up. Their research group have also included pretrained models for cats, cars, and bedrooms in their repository that you can immediately use.
Each time you refresh the site, the network will generate a new facial image from scratch from a 512-dimensional vector.
StyleGAN was trained by the NVIDIA Research Projects team using the CelebA-HQ and FFHQ datasets for an entire week using 8 Tesla V100 GPUs according to Rani Horev’s explanation.
While it is clear that the use of StyleGAN to achieve these results is nothing short of impressive, the fact that every generated face also has some little details that don’t line up exactly and break the overall symmetry, rendering the face even more eerily realistic looking, is even more remarkable.
If you want to see how StyleGAN operates behind the scenes to generate these fake faces of people that have never existed, watch the video below.
While thispersondoesnotexist.com showcases what one can achieve using a StyleGAN version trained to work with human faces, other people have made their own StyleGAN models and trained them to generate anything from font variations, psychedelic graffiti, and cat faces. Be warned though, those cat faces are … something else.
#LatentFonts as the next iteration of #VariableFonts? Tried #StyleGAN on a 50k fonts dataset collected by @fulhack – Now I'm stuck watching this loop pic.twitter.com/UKCexhJKiF
— Cyril Diagne (@kikko_fr) February 13, 2019
Read the original article over at BleepingComputer.com.