
NVIDIA, regularly impressive to the public with new neuralities intended for working with images, demonstrated a new development, thanks to which everyone can feel like a real artist.
According to TechCrunch , an algorithm called Gaugan, the consonant of the artist of the Gogen field, turns primitive sketches of landscapes into realistic images. Judging by the demonstration video, the user can choose a tool (“pencil”, “brush” or “filling”), as well as a type of object that he wants to portray. It can be water, rock, wood, grass, cloud, stone, sand or other object.
After the sketch on the original image, the program processes it, turning simple scribbles into a landscape photo. At the same time, the algorithm takes into account many details - in the final image, not only the necessary objects appear, but also their shadows and reflection in water. And although in the current form the program sometimes makes mistakes distinguished on the contours of objects, its capabilities are very impressive, and in Nvidia they promise to improve the work of Gaugan.
To teach the algorithm, the creators used approximately 1 million images from the Flickr photo hosting. NVIDIA emphasizes that all these photos were allowed to use for such purposes. Thanks to training on this dataset, the algorithm can generate hundreds of thousands of different objects and their combinations.
Soon, the Gaugan demo version should appear as part of the NVIDIA AI PlayGround project: with its help, anyone will be able to test this and other technologies of the company to create and process images in action, writes N+1 .
Recall that in February NVIDIA showed the neural network Stylegan, which generates realistic faces of non -existent people.