Nvidia's R&D team showcased GauGAN at GTC 2019, an innovative neural network solution that greatly simplifies the work of designers and artists. There is no longer any need to draw photorealistic landscapes with your own hands, because Nvidia's technology is capable of transforming images into images that are almost indistinguishable from reality with just a few sloppy strokes.
The technology is named after the French post-impressionist Paul Gauguin and is most of all similar to the new generation of Paint. The user requires the very minimum of effort - select in the menu options for what he wants to depict in the picture (at the moment there are twenty options available: from sky to gravel) draw the shape of the object, select a color and in a matter of seconds the lurid workpiece is digitized into picturesque landscapes.
It is important to note that GauGAN takes into account the context of the picture and if the user, for example, draws a mountain near a reservoir, then a realistic reflection of the mountain and the entire surrounding space will appear on the water surface. The presence of several customizable changes makes it even easier to work with the editor, with which you can significantly modify the picture in a few clicks. For example, to change the time of day or even the season in the program, because of which the grass will be covered with snow, the trees will lose their foliage, and the clouds in the sky will acquire a lead shade familiar for winter.
Intelligent completion of the landscape in GauGAN is possible thanks to the use of generative adversarial neurons and sequential training of the program based on more than 1 million photographs received from the photo hosting Flickr.
As acknowledged by Nvidia employees, GauGAN is being developed for the sake of making everyone feel like an artist and, at the same time, becoming a convenient tool in the hands of specialists. GauGAN capabilities can be used by both architects and designers, developers of games and projects for virtual reality. The functionality of GauGAN, as conceived by the developers, should significantly speed up the development and editing of products. In addition, the developers do not exclude that in the future the program will learn how to work with people, cars, animals and any other objects, because the most important requirement is to "feed" the neural network an extensive pack of photos on the required topic.
At the moment it is impossible to download GauGAN in the public domain, and Nvidia is not ready to release the project, since it still takes time to polish the technology. But now, all attendees of the GTC 2019 conference in California can try the GauGAN demo.
The Topic of Article: Nvidia has unveiled the GauGAN neural network capable of turning a rough sketch into a photorealistic landscape.