Sign In

Communications of the ACM

ACM TechNews

AI Artist Conjures Up Convincing Fake Worlds From Memories


An image generated by an imaginative neural network.

Stanford University and Intel researchers have developed a neural network that can create realistic images based on real-world photos.

Credit: Stanford University and Intel

Researchers at Stanford University and Intel have developed a neural network that is capable of creating realistic photos based on real-world images.

The team says the artificial intelligence (AI) system works from rough layouts that tell it what should be in each part of the image.

The new technique also could eventually be used to create game worlds that resemble the real world, says Stanford professor Qifeng Chen.

The first step is for the system to process a photo of a real street it has not seen before, but that has been labeled so the AI can identify objects. The program then uses the layout as a guide to generate a completely new image.

The Stanford system was trained on 3,000 images of German streets, so when it comes across a part of the photo labeled "car," it draws on its existing knowledge to generate a car in the new photo.

From New Scientist
View Full Article

 

Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


 

No entries found