Face Depixelizer using Machine Learning
Using the face depixelizer, I used AI to turn an anime character into real a life character by giving their faces a more human touch. Some results are quite funny and cursed.
The idea behind it is based on PULSE: A Self-Supervised Photo Upsampling via Latent Space Exploration of Generative Models.
PULSE borrows the output from StyleGAN, a generative adversarial network to produce the results that we see here. The model takes quite a while to create an output, and the output didn't vary much based on pixelation or blurring.
Hope you enjoyed the video as much as I did playing around with this technology! Definitely worth the wait for the output.
Notebook - https://colab.research.google.com/git...