NVIDIA Designs Prototype Software To Convert Rough Doodles Into Lifelike Landscapes

The new AI software developed by NVIDIA allows users to paint any texture they think and imagine of and allows artists to demonstrate their artwork.

Generative Adversarial Network (GAN) allows users to generate artwork that look similar to authentic and realistic landscapes. NVIDIA used this artificial intelligence model which allows the users to draw simple and rough doodles and fill the outlines with naturalistic textures.

Bryan Catanzaro, Vice president of applied deep learning research at NVIDIA said, “It’s much easier to brainstorm designs with simple sketches, and this technology is able to convert sketches into highly realistic images.” The AI software has three different tools to acquire a realistic picture: a pencil to draw rough doodles, a pen to highlight the pictures and a paintbucket to fill in colors and textures. The entire processing of the image does not mean putting the puzzles of the picture together but actually creating each pixel to knit the image. Generator network has the ability to create basic elements and then presenting it to the discriminator which in turns tells the former upon how to improvise the picture using Flickr as a reference. Interestingly, GAN changes the other segments of the picture to beautify the picture when even one of the element is improvised.

The software is built in a manner that if the grass is converted into snow, it will know that the appearance of the sky is also to be changed. It knows when to add water reflections and artistic lines to make it look real. However, the software does have a few limitations. Pictures of buildings and furniture are difficult to achieve as they look unrealistic. All the other elements in the picture can impressive, but the results cannot resemble an actual realistic photograph. Nevertheless, such tools can help architects, game developers and urban town planners.