The New AI of Nvidia Converts Sketches Into Photo-realistic Landscapes

27 Mar 2019 Wednesday
The New AI of Nvidia Converts Sketches Into Photo-realistic Landscapes
Ağahan  Çelik
Ağahan Çelik Editor

The software uses Flickr images during the conversion process

Nvidia presented an artificial intelligence (AI) that takes the processing of images to a really surprising level. Do you remember the scribbles you used to do in Paint for fun? The software, named GauGAN, is capable of converting those simple drawings into authentic photo-realistic landscapes. It is important to mention that it is still a prototype, so they have a margin to improve their performance.

The California-based company uses the generative adversarial network (GAN) to offer a “smart” brush. It is possible, for example, to draw a tree on a hill and the AI will be responsible for providing the appropriate textures. In the following video you can see in detail the operation of the application.

Nvidia's GauGan software turns sketches into paintings

The company's engineers made use of a million Flickr images - with a Creative Commons license - to train the neural network. Artificial intelligence analyzes the elements that appear in photographs and compares them with real-life objects.

Of course, GauGAN requires high-performance hardware to give instant results. They use the Titan RTX GPU, one of the highest performance graphics cards on the market; its price is $2,499. Nvidia mentions that its software can work with almost any existing GPU, however, the time to generate the landscapes will depend on the power of the computer.

Once they have made the necessary improvements, GauGAN’s technology can be applied in the development of video games to create "more immersive environments".