AI paints lifelike cityscape from memory alone

Researchers from Stanford University in California and Intel have created an artificial intelligence (AI) program that can paint a cityscape from memory.

Alphr says that the system, developed by PhD student Qifeng Chen and Intel, uses "thousands of reference images" to learn where cars, pavements and road signs are located in a city before painting an image depicting a real-world environment.

The image, however, "technically doesn't exist at all" and is simply formed by the AI program remembering what it has learned from "3,000 images of German streets" and painting its own interpretation of a city from memory alone, the website says.

Subscribe to The Week

Escape your echo chamber. Get the facts behind the news, plus analysis from multiple perspectives.

SUBSCRIBE & SAVE

Sign up for The Week's Free Newsletters

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

Sign up

To help the program paint the image, TheNextWeb says researchers create a "paint by numbers" template that consists of basic coloured blocks that outline a conventional cityscape.

The program could have significant benefits for video game creators, the site says, as it could be used to help open-world titles such as Grand Theft Auto produce more realistic cities.

But there's room for improvement. NewScientist says the images are blurry because the network "isn't able to fill in all the details we expect in photos." An improved version of the AI program is already in development that's expected to improve the quality of the images.

Chen told the website that the program could be used with voice commands in the future, as "it'd be great if you could conjure up a photorealistic scene just by describing it aloud".