Jun 23, 2021 20:00:00

GTA 'GAN Theft Auto' generated by neural network is released

Using a hostile generation network (GAN) that enhances the accuracy of data learning by competing with two neural networks, it automatically generates a game like 'Grand Theft Auto V (GTA V)'. 'GAN Theft Auto ' was created and released by sentdex of YouTuber.

GitHub --Sentdex / GANTheftAuto
https://github.com/sentdex/GANTheftAuto/

Playing a Neural Network's version of GTA V: GAN Theft Auto --YouTube


GAN Theft Auto is based on 'GameGAN ', a game generation AI created by NVIDIA. GameGAN, which is the basis of GAN Theft Auto, is an AI that can reproduce Pac-Man's game itself with just the game play movie without preparing a game engine or code.

NVIDIA's AI reproduces 'Pac-Man' without a game engine just by watching 50,000 play videos --GIGAZINE

by WishItWas1984

GAN Theft Auto specializes in reproducing GTA V. sentdex created a mod for driving a car in GTA V to collect learning data. This mod helps you collect various environmental data in GTA V by automatically entering steering commands and changing speeds. The following movie shows how to actually collect data in GTA V using MOD.

GAN Theft Auto data collecting demo --YouTube


sentdex hasn't published this mod, but he does provide a sample dataset collected using the mod. If you find interesting results with AI models other than GAN Theft Auto, please feel free to contact us to provide a complete dataset.

GAN Theft Auto is open source and published on GitHub.

GitHub --Sentdex / GANTheftAuto
https://github.com/sentdex/GANTheftAuto/

It is also possible for the player to actually drive a car in GAN Theft Auto output by GAN as a GTA V-like game. GAN Theft Auto uses PyTorch , Python's open source machine learning library.

GAN Theft Auto supports 4x upscaling and upsampling to improve the visibility of the output data. The actual improvement in image quality is as follows. The image on the left is before upscaling and the image on the right is after upscaling.