GauGAN: NVIDIA AI turns sketches into photorealistic landscapes


We continue with the news of the GPU Technology Conference after the announcement of the single-deck computer Nvidia Jetson Nano of $ 99 dedicated to the implementation of applications in artificial intelligence for developers, researchers and hobbyists.

In this same GTC 2019, Nvidia, the global provider of processors and graphics chips revealed an image maker animated by artificial intelligence. The software called GauGAN by its designers, provides an overview of the possibilities offered by Nvidia neural network platforms.

This AI builds on learnings from the Pix2Pix system introduced last year that can represent virtual worlds, said Nvidia's vice president of applied deep learning research Bryan Catanzaro, but Pix2Pix can't paint landscapes because doing so leaves artifacts in the resulting image.

GauGAN is designed to make a sketch and turn it into a photorealistic image in seconds. GauGAN offers three tools: a paint bucket, a pen, and a pencil.

The GauGAN Demonstration in the current edition of the GPU Technology Conference follow the launch, in the middle of the previous month, from a site that shows portraits of human faces generated by artificial intelligence.

It should be noted that, at the end of the previous year, the company had already presented an artificial intelligence capable of generating human faces of a worrying reality.

GAN concept

The common denominator of these initiatives with GauGAN software is the GAN concept.

A GAN is a generative model in which two networks compete in a game theory scenario.

The first network is the generator, generates a sample (for example, an image), while its adversary, the discriminator, tries to detect if a sample is real or if it is the result of the generator.

Learning can be modeled as a zero-sum game. These computer programs compete millions of times to improve your imaging skills until they have the ability to create complete images.

Simply put, GAN means that two networks work against each other.

It is first fed raw data that is decomposed. From these, trat create an image. Lthen send it to another network that, it only has real photos or images in its database. This second network will make a judgment of the image and will inform the first.

If the image does not look like the expected result, the first algorithm resumes the process. If there is a match, you are informed that you are on the right track and you end up understanding what a good image is.

Is that how it works GauGAN

Once you are sufficiently trained, you can produce images on the chain. According to data published by Nvidia, the discriminator running in the background of the GauGAN software has a database of a million images of nature.

GauGAN could offer a powerful tool for creating virtual worlds. Even in this limited demo, it is clear that the software built around these skills it would appeal to everyone from video game designers to architects to casual gamers.

With an artificial intelligence that understands what the real world looks like, these professionals could better prototype their ideas and make quick changes to a synthetic scene.

The company has no plans to release it commercially, but could soon launch a public trial to allow anyone to use the software.

Through the GauGAN software demo, Nvidia highlights the positives of use of technologies that are based on the GAN, But it must be said that this set of techniques can also be used for sinister purposes.

Deepfakes (computer generated images superimposed on other or existing videos) are part of this lot and are trusted by malicious third parties to spread false news and hoaxes.

Nvidia maintains the AI ​​Playground online platform. It lists the projects the company is launching on in terms of artificial intelligence and internet users have the opportunity to launch demos.

The content of the article adheres to our principles of editorial ethics. To report an error click here!.

A comment, leave yours

Leave a Comment

Your email address will not be published.



  1. Responsible for the data: Miguel Ángel Gatón
  2. Purpose of the data: Control SPAM, comment management.
  3. Legitimation: Your consent
  4. Communication of the data: The data will not be communicated to third parties except by legal obligation.
  5. Data storage: Database hosted by Occentus Networks (EU)
  6. Rights: At any time you can limit, recover and delete your information.

  1.   Marcela said

    How cool