The sprawling cityscapes of Grand Theft Auto V have been given an AI-powered makeover that makes the game’s online world look almost photorealistic.
Researchers at Intel Labs transformed the graphics by training convolutional networks on real-world images.
The network then applied the style of the training data to rendered images from the game.
In one example, the team trained the model on the Cityscapes dataset, a collection of urban street scenes recorded primarily in Germany.
This generated game graphics with smoother asphalt on the roads, glossier car paint, detailed reflections on windows, and greener grass and hills.
It looks great, but the color schemes of urban Germany aren’t an ideal match for GTA V’s LA-inspired setting. But the researchers achieved a more visually vibrant recreation by training the network on the geographically diverse Mapillary Vistas dataset.
I personally enjoyed the Deutsch GTA, but both the transformations looked pretty stunning to me. You can judge them for yourself by watching the video at the top of this article.
You’re not likely to get a playable mod based on the system any time soon, but if it ever gets productized, it could make the thrill of virtual joyriding more realistic than ever.
Greetings Humanoids! Did you know we have a newsletter all about AI? You can subscribe to it right here.
Get the Neural newsletter
Greetings Humanoids! Did you know we have a newsletter all about AI? You can subscribe to it right here.Follow @neural