This Halloween Turn Yourself Into A Zombie With This AI Tool Using StyleGAN2

0
726
Source: https://makemeazombie.com/

Many might be interested in an AI tool that changes your picture into a zombie with Halloween season upon us. NVIDIA in early 2019 open-sourced StyleGAN (Style Generative Adversarial Network). This GAN architecture extension can generate impressively photorealistic images while enabling user control over image style. With a new release this year, NVIDIA improved StyleGAN2 with redefined state-of-the-art image generation, inspiring fun and creative pursuits with faces.

StyleGAN tech-inspired the last month’s viral Toonify Yourself website, created by a couple of independent developers. It turns selfies into big-eyed cartoon characters. A Nebraska-based developer Josh Brown Kramer has taken facial image transfer tech to a new height, building a zombie generator. He has set up a public website to upload any pics and turn it into a zombie using the generator for free.

Advertisement

Kramer explained that he first transfer-learned a StyleGAN2 zombie generator; after that, he created a hybrid StyleGAN2 model. He was inspired by the Reddit post-Cross-Model Interpolations Between 5 StyleGanV2 Models – Furry, FFHQ, Anime, Ponies, and a Fox Model. The original human image generator is the first layer of the model, and the zombie generator is the last. Finally, drawing from the Yandex and Moscow Institute of Physics and Technology paperStyleGAN2 Distillation for Feed-forward Image Manipulation, Kramer used fifty thousand image pairs to learn how to map between the image pairs.

https://syncedreview.com/2020/10/27/ai-halloween-avatars-stylegan2-generator-reveals-your-inner-zombie/

The two differences between Toonify Yourself and Make Me A Zombie are as follows:

  • Use of crappify by the zombie generator for input data enhancement, introducing resizing and compression artifacts.
  • Focusing by the Toonify Yourself generator on its cartoon images’ texture, while the zombie model emphasizes the original image’s shape and orientation.

Trained on a hand-picked zombie dataset from Pinterest and Google, the system comprises 300 images of people in zombie makeup and Halloween masks. By adjusting the learning rate and the batch size, Kramer trained the model at a 1024×1024 size on his personal Nvidia 2080 Ti GPU in a day (approximately).

This has now become a trend for people of all age groups. Check out the Make Me A Zombie website to learn more and even create your zombie images.

Related Paper: https://arxiv.org/pdf/2003.03581.pdf

Source: https://syncedreview.com/2020/10/27/ai-halloween-avatars-stylegan2-generator-reveals-your-inner-zombie/

Tool: https://makemeazombie.com/

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.