One of the most impressive aspects of 'Grand Theft Auto V'(' GTA V ') is what San Andreas in the game looks like in Southern California in real life. Now, a new machine learning project (machine learning) Of the Intel Labs can allow photo-realistic graphics on games in a near future.

Call "Enhancing Photorealism Enhancement", The technology can take realism in a photorealistic way to" disturbing ", according to the group of researchers from Intel working on the project. Stephan R. Richter, Hassan Abu Alhaija and Vladlen Kolten have created a look that has unmistakable similarities to the types of photos you can take casually through a car window. (see video below).

advertising

With the use of artificial intelligence (AI), the tool analyzes all frames of the game (in the case of the video above, 'GTA V'), comparing them afterwards with real photographs from a database of several cities. Then, choose the images that most resemble the scenes in the game and apply improvements to each of the frames, making it more realistic.

Even though it is not clearer than a game common, the image enhanced by Enhancing Photorealism Enhancement is able to confuse even a well-trained eye. The devices used include smudges and dirt that are normally difficult to reproduce in 3D graphics or in 8K resolution, in addition to brightness and reflections strategically positioned by the AI.

And this is so, according to the researchers, because the improvements here "go beyond what other photorealistic conversion processes" are capable of, since Intel technology is also able to integrate geometric information from the 'GTA V' itself. The so-called “G-buffers”, as the trio calls them, can include data such as the distance between objects in the game and the camera, in addition to the quality of textures, such as the brightness of carros.

Read more:

In short: technology can only work with what is received. Therefore, by improving the quality of what is being fed to it and the way it is being fed, the results can be improved dramatically.

Even though it is still under development, the improvement with regard to the realism of the scenario is already clear. And thanks to the neural networks next to machine learning, in Enhancing Photorealism Enhancement can produce images at an “interactive speed”. That is, at least in theory, it is possible that the tool can act in real time and improve on its own, through updates.

As it is still an experimental project, it is not known whether the image enhancement technology will be interested in its application in the near future. In addition to the released video, most of the data and findings from the Intel Labs team is available at project page, in a much more detailed document.

Sources: The Verge e Kotaku

You’ve already watched our videos on YouTube? Subscribe to our channel!