Neural networks help create detailed picture of Doomguy

A Redditor created a high-resolution image of Doomguy using a range of face generators, including Nvidia’s StyleGAN. After several iterations, the user managed to obtain a relatively realistic face.

Doomguy’s scarred face in the original 1993 version of the game is one of the iconic effigies in gaming history. Redditor zergling103 wondered what the robber would look like with a less pixelated look.

He therefore already used the applications FaceApp, Waifu2x and Gimp a year ago to obtain a first, somewhat more detailed result. He recently fed that image along with the original image to StyleGAN. This face generator, developed by Nvidia’s NVlabs, uses generative adversarial networks, an architecture in which neural networks play a game to improve their results.

Zergling103 manually adjusted the result and made the face a bit smaller, for example, after which he had the image processed by StyleGAN again. He also gave the original a smile with teeth to see what result the neural networks would come up with.