A paper published on ArXiv, a repository of non peer-reviewed research, reveals that new tech created by a team from the Korea Advanced Institute of Science and Technology (KAIST) School of Electrical Engineering has the ability to create pretty accurate photos of highly pixelated images.
Naturally, we’ve already started to abuse the tech.
Deokyun Kim et al. write that “the main challenge of face Super-Resolution (SR) is to restore essential facial features without distortion.” In order to create accurate renditions, Kim’s team broke the process up into steps, with each successive step “producing output with a progressively higher resolution.”
We propose a compressed version of the state-of-the-art face alignment network (FAN) for landmark heatmap extraction. With the proposed FAN, we can extract the heatmaps suitable for face SR and also reduce the overall training time.
— Kim et al.
The researchers were able to successfully train their machine-learning tool to recognize faces among blurry pixels and build a close rendering.
Then, as it does, the Internet found another use for the facial SR technology: creating photorealistic versions of 16×16 pixel emojis.
The results are horrifying. When fed into the machine learning model, a pepperoni slice becomes a gooey pile of cheese with human mouths, teeth and all, in place of pepperonis. The nervous smile emoji, meanwhile, is transformed into a creepy overstretched face straight out of the Uncanny Valley. See for yourself:
“Because this model is trained specifically to look for facial landmarks, it will take any excuse to draw eyes and nostrils on a pixel,” writes Jonathan Fly, an AI expert and the man behind the distorted emojis, on his blog.
“I also went to some lengths to encourage the model to make aggressive guesses about facial features,” Fly writes, explaining why some of the emojis are especially terrifying.
The only thing worse than these disturbingly humanoid emojis is probably something like this disembodied animatronic head.