AI paints a realistic new world
Qifeng Chen from Stanford University has developed a deep learning algorithm synthesizing very realistic images. The neural network is capable of imagining realistic urban landscapes that are actually computer-generated. It was fed with 30,00 photos of German streets. The different layers of the photos (tree, road, traffic signs, cars, etc.) were identified so that the AI can learn to imagine a lifelike new world from these “memories”.
For now, the created images are still a little blurry and dream-like or similar to moving oil paintings, but Qifeng Chen’s ultimate goal is to use this AI to build complete virtual worlds very quickly, for example for video games. This innovation questions whether digital intelligence can have imagination, or even and artistic ambition, even if it remains limited by a database.
Spotted in 2018
- United States