[ad_1]
People across the world will soon put on scary makeup, terrifying masks and spooky costumes in honor of Halloween, but an algorithm is capable of transforming you into a flesh-eating zombie in just seconds.
The website, called MakeMeAZombie,’ stems from Toonify’s ability to transform portraits into fun-loving cartoons, but has been redesigned to make users look like the undead.
The algorithm was trained on a sample of 300 pictures from the internet of people wearing zombie masks and makeup, which were then combined with images of traditional human faces to teach it how to map the ghoulish features.
The resulting image shows the person with angry, beady eyes, a mouth of rotten teeth and skin that appears to be decaying.
Scroll down for video
An algorithm is capable of transforming you into a flesh eating zombie in just seconds, just as it made US President Donald Trump look like the undead
MakeMeAZombie is the brainchild of Josh Brown Kramer, who is a senior data science consultant for Lincoln-based medical AI developer Ocuvera, and has been working on the system for nearly one year.
The project required Kramer to pair 50,000 images of artificial intelligence-generated human faces with corresponding zombie faces.
‘I initially started trying to turn people into dogs,’ Kramer told Silicon Prairie News, describing his project’s origin.
‘Then a friend of mine had the idea of doing zombies, which just sounded much cooler, and like it would be a lot of fun for Halloween.’
The website, called MakeMeAZombie,’ stems from Toonify which transforms portraits into fun-loving cartoons, but has been redesigned to make users, or even Vice President Joe Biden, look like the undead
The algorithm was trained on a sample of 300 pictures from the internet of people wearing zombie masks and makeup, which were then combined with images of traditional human faces to teach it how to map the ghoulish features, such as the decaying features shown on Kim Kardashian West
He gathered a zombie data set of 300 images of people representing the living corpses from Google and Pintrest.
These images were then fed into the neural network StyleGan2, which is a system designed by NVIDIA – an American-based AI computing firm.
And finally, Kramer taught the algorithm how to map faces using 50,000 pairs of human and zombie faces.
‘I dumped 50,000 pairs of images — the first from the human StyleGAN2 generator, and the second with the same latent space representation, but passed through the zombie generator,’ Kramer explained on Reddit.
‘I then used Pix2PixHD to learn a mapping from the pairs.’
AI has been used to transform traditional images into nightmarish scenes for years, such as MIT’s innovation that generated creepy images from famous sights around the world.
The resulting image shows the person with angry, beady eyes, a mouth of rotten teeth and skin that appears to be decaying
Users simply upload a selfie or another image, such as the one of Elon Musk shown here, and MakeMeAZombie does the work in just seconds
The algorithm created by the team is making photographs of famous landmarks appear like something out of a horror film.
‘We use state-of-the-art deep learning algorithms to learn how haunted houses, or toxic cities look like,’ the researchers said.
‘Then, we apply the learnt style to famous landmarks and present you: AI-powered horror all over the world!’
The two main techniques used in the project, style transfer and generative adversarial networks, were published in papers last year.
The network typically consists of 10 to 30 stacked layers of artificial neurons and each image is fed into the input layer, which then talks to the next layer, until eventually the ‘output’ layer is reached.
The network’s ‘answer’ comes from this final output layer.
In doing this, the software builds up a idea of what it thinks an object looked like.
‘In the ‘generative adversarial network,’ part of the network will attempt to fool the other part by inventing fake data, which will be mistaken for training data.
In creating a network that works against itself, researchers believe it will eventually learn to be more precise in its output.
‘We use state-of-the-art deep learning algorithms to learn how haunted houses, or toxic cities look like,’ the researchers said.
‘Then, we apply the learnt style to famous landmarks and present you: AI-powered horror all over the world!’
[ad_2]
READ SOURCE
