Nvidia AI turns sketches into photos-realistic landscapes in a matter of just a few seconds

Today at the Nvidia GTC 2019 The company revealed an impressive image creator. With the help of generative adversarial networks, users of the programme have only a few clicks in them to draw images that are almost real-world. The programme will instantly transform two lines into an amazing mountaintop sunset. It’s MS Paint for the AI age.

The programme is dubbed GauGAN The software is an example of what’s possible using neural networks on Nvidia platforms. It’s designed to create an image that humans paint, and the aim is to sketch a sketch and transform it into a photorealistic photograph in just a few minutes. In the early tests, it worked exactly as it claimed.

Gaugan comes with three tools, including a paint bucket, pen, and pencil. At the bottom of the screen, you’ll find an array of objects. Choose the cloud object, and trace a line using the pencil. The software will create an image-realistic cloud. However, these aren’t images. GauGAN generates results that are unique in the context of the input. Draw a circle, then fill it with paint, and the programme will create fluffy summer cloud formations.

Users can make use of in-built tools for drawing the form of a tree. This will create trees. Create a straight line, and it will result in a barren trunk. Draw a bulb on the top, and the programme will fill it with leaves, creating an entire tree.

GauGAN can be multimodal as well.If two users make the same sketch using the same parameters, the random numbers that are built into the project guarantee that GauGAN produces different outcomes.

In order to get instantaneous outcomes, GauGAN has to run on a Tensor computer platform. Nvidia showed the programme on its RDX Titan GPU platform, which allowed it to generate results in real-time. The person who was running the demo was in a position to draw an outline, and the programme immediately produced results. But Bryan Catanzaro, VP of Applied Deep Learning Research, said that with a few modifications, GauGAN can run on almost every platform that includes CPUs, although the results may take a couple of seconds to show.

In the demonstration, there are boundaries that separate objects. They aren’t perfectly drawn, and the team that is behind this project says they will work on it. There’s a faint line between two objects. Nvidia claims that the result is realistic; however, when examined, the results don’t hold up. Neural networks have issues with both the objects they were taught on and the tasks they were designed to perform.The project aims to close the gap.

Nvidia altered around 1 million pictures from Flickr to help train its neural network. The majority of them were taken directly from Flickr’s Creative Commons, and Catanzaro confirmed that the company is only using images that have been granted permission. The company claims that this software can synthesise thousands of objects and their relationships to other objects that exist in the real world. It is possible to change the seasons in Gawanak, and the leaves on the trees will fall off.If there’s a pond right in the vicinity of a tree, the tree will reflect in the water.

Nvidia will announce the white paper on its website today. Catanzaro said that the paper was previously approved for CVPR 2019.

Catanzaro believes this programme will be made available through Nvidia’s latest AI Playground, but he acknowledges that there’s some work the company has to complete to make it happen. The tools he’s describing are being utilised within video games in order to build immersive environments However, Nvidia has not specifically developed software for this purpose.

It’s easy to complain about the ease with which software can be used to create fake images to serve nefarious ends. Catanzaro is of the opinion that it’s a crucial issue and says it’s more than just one person or company. “We care about this a lot because we want to make the world a better place,” he stated while stating that it’s an issue of trust, not a technological issue, and that we, as a community, have to deal with the issue in a different way.

In this demonstration, it’s evident that software that is built around these capabilities would be appealing to anyone from video game designers and architects to casual gamers. The company doesn’t have any plans to launch it for sale, but it may soon be releasing a trial version for anyone to try the program.