Monday , November 29 2021

Nvidia created the first game demo using the & # 39; graphics generated by CA



[ad_1]

The recent boom in artificial intelligence has produced impressive results in & # 39; environment somewhat surprising: the world of image and video generation. The latest example comes from the Nvidia chip designer, today published research showing how visual generated by the CA can join & # 39; traditional engine of videogames. The result is & # 39; hybrid graphics system that can & # 39; used once & # 39; computer games, movies and virtual reality.

"It's a new way to render video content using the & # 39; deep learning," said vice-president & # 39; Nvidia of & # 39; profound learning application, Bryan Catanzaro. The Verges. "Obviously Nvidia takes a lot of thinking about the graphics generation [and] we are thinking about how AI will revolutionize the field. "

The results of the work of & # 39; fotorealistiċi and Nvidia are showing visual tamper trademark found in & # 39; many images generated by AI. Nor are entirely new. F & # 39; of & # 39; research paper, the company's engineers explain how they constructed a number of & # 39; Existing methods, including influential system & # 39; pix2pix called open source. Their works use type & # 39; neural network known as contentious generating network, or GAN. They are used by & # 39; broadly & # 39; in the generation & # 39; AI image, including the creation of & # 39; photo CA recently sold by Christie & # 39; s.

But Nvidia has introduced a number of & # 39; innovations, and one product & # 39; this work, he says, is the first video game demo b & # 39; generated by AI machines. It's simple driving simulator where players navigate a few blocks of the city & # 39; space generated by the AI, but can not leave their car or interact with the world. The demo is powered using the & # 39; only one GPU – a remarkable achievement for such advanced work. (Although undoubtedly GPU is the best run company & # 39; $ 3,000 Titan V, "the GPU PC more b & # 39; 'health ever created" and one typically used for processing & # 39, rather than advanced simulation games).

The system & # 39; generate graphics using nVidia few steps. First, researchers must collect training data, which in & # 39; in this case taken from open source datasets used for autonomous driving research. This footage is then segmented, which means that each frame is divided into & # 39; different categories: sky, cars, trees, roads, buildings, etc. generating network issue then trained on this segmented data to generate new versions of & # 39; these items.

Then, engineers created the basic topology of the virtual environment using the & # 39; traditional gaming machine. F & # 39; in this case the system was Unreal Engine 4, Engine used for such popular titles Fortnite, PUBG, Gears of War 4, And many others. By & # 39; this environment as a framework, the algorithms & # 39; profound learning then generate graphics for each different category of & # 39; & # 39 item in real-time, and merging them with models of the game machine.

"The structure of the earth is traditionally created," Catanzaro explains, "the only thing that generates AI is styling." He adds that the demo itself is basic, and it was done by one engineer. "It's more proof of concept than a game the pleasure to play."


Comparison & # 39; images generated by AI. On the left is the segmentation map; pix2pixHD top right; COVST bottom left; down on the right, the system & # 39; Nvidia, vid2vid.
Credit: Nvidia

To create this system, the engineers & # 39; Nvidia had to work in & # 39; number & # 39; challenges, the biggest of which was the permanence of the object. The problem is whether the algorithms & # 39; deep learning are generating the graphics for the world b & # 39; of & # 39 rate; 25 frames per second, how to keep items that look the same? Catanzaro says that the problem meant that the initial results of the system were "hard to look" as colors and textures "changed every frame."

The solution was to give the short-term memory system, so it can & # 39; compare each new frame with & # 39; what went before. It tries to predict things like motion in & # 39; these images, and create new frames that are consistent with & # 39; what's on the screen. All this is expensive computation expensive, so the game runs only in & # 39; 25 frames per second.

The technology is very much in the early stages, emphasizes Catanzaro, and x & # 39; likely to be few & # 39; years unless the graphics generated by the AI ​​seen in consumer titles. He compares the situation to the development of ray tracing, the warm current techniques in graphics rendering when individual beams & # 39; light generated in & # 39; real-time to create realistic reflections, shadows and opacity f & # 39; virtual environments. "The first interactive ray beam interactive happened a long time ago, but failed to get in games until a few weeks ago," he says.

The work is not potential applications in & # 39; other areas & # 39; research, though, including robotics and automatic driving cars, which can & # 39; used to generate & # 39; training environments. And it & # 39; manifest itself in consumer products earlier, although in & # 39; a more limited capacity.

For example, this technology can & # 39; used in & # 39; of & # 39 system; Hybrid graphics, where the majority of the game provided by & # 39; traditional methods, but AI is used to create images of & # 39; people or objects. Consumers can capture the footage themselves using the & # 39; smartphones, then upload & # 39; this data to the cloud where jikkopjawha algorithms learn and put in & # 39; games. This makes it easier to create avatars that resemble such players, for example.

However, this type of & # 39; technology raises some obvious questions. F & # 39; in recent years, experts have become more concerned about the use of & # 39; deepfakes generated by CA disinformation and propaganda. Researchers have shown that it is easy to generate a fake video & # 39; politicians and celebrities to say or do things that were not, powerful weapon in & # 39; wrong hands. By pushing & # 39; to advance the capabilities of & # 39; this technology and publishing its research, Nvidia x & # 39; likely to contribute to this potential problem.

The company, though, says that this is not a new issue. "Can & # 39; [this technology] used for creating & # 39; content that is misleading? Yes. Any technology for rendering can & # 39; used to do it, "says Catanzaro. He says Nvidia is working with & # 39; partners for ways & # 39; research to discover AI fake, but ultimately recently Problem of & # 39; bad information is "a matter of & # 39; trust ". And like many issues & # 39; trust before, have to be resolved by & # 39; many ways, not just technology.

Catanzaro says that technology companies like Nvidia can take so much responsibility. "You have the power company responsible because it created electricity to empower the computer to make fake video?" Calls.

And finally, for Nvidia, push & # 39; hereinafter the graphics generated by the CA has an obvious benefit: it will help the company sell more hardware. Since the detonation of teaching left in early 2010, the price of the stock & # 39; Nvidia increased as it seemed obvious that their computer chips were ideally suited for research and the development of learning machine.

So the AI ​​revolution in computer graphics is good for the company's income? Certainly will not hurt & # 39 ;, Catanzaro laughs. "Anything that increases our ability to generate graphics that are more realistic and compelling think it is good for the end of the line Nvidia."

[ad_2]
Source link