焦點

【】

字号+作者:囫圇吞棗網来源:知識2025-04-04 12:54:49我要评论(0)

The image above looks like a collage of photographs, but in fact, it's been generated by an artifici

The image above looks like a collage of photographs, but in fact, it's been generated by an artificial intelligence. And as real as they may look, the people in the image aren't actual humans.

In a new paper (via The Verge), a group of Nvidia's researchers explain how they've created these images by employing a type of AI, called generative adversarial network (GAN), in novel ways. And their results are truly mind-boggling.

SEE ALSO:Waymo launches self-driving taxi service, but caveats abound

The paper is titled "A Style-Based Generator Architecture for Generative Adversarial Networks" and signed by Tero Karras, Samuli Laine and Timo Aila, all from Nvidia. In it, the researchers show how they've redesigned the GAN's architecture with a new approach called "style-based design."

"Our generator thinks of an image as a collection of "styles."

"The new architecture leads to an automatically learned, unsupervised separation of high-level attributes (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair)," the paper says.

In layman's terms, after being trained, the GAN produced images that are pretty much indistinguishable from photographs of real people, completely on its own.

Mashable Light SpeedWant more out-of-this world tech, space and science stories?Sign up for Mashable's weekly Light Speed newsletter.By signing up you agree to our Terms of Use and Privacy Policy.Thanks for signing up!

"Our generator thinks of an image as a collection of "styles," where each style controls the effects at a particular scale," the researchers explain in a video accompanying the paper. These styles are attributes such as pose, hair, face shape, eyes and facial features. And researchers can play with these styles and get different results, as seen in the video, below.

It's not just people that GAN can create in this way.

In the paper, the researchers use the GAN to create images of bedrooms, cars and cats.

Mashable ImageCredit: Nvidia/Arxiv.org

Amazingly, the concept of GANs was introduced just four years ago by researchers from the University of Montreal.

Check the image from that paper below to see how much progress has been made since then.

Mashable ImageCredit: Universite de Montr EAL/arxiv.org

It's easy to see this technology used in the creation of realistic-looking images for marketing or advertising purposes, for example. But it's just as easy to imagine someone using it to create fake "evidence" of events that never happened in order to promote some agenda.

At the speed this tech is progressing, it soon might be impossible to tell whether you're looking at a real photograph or a computer generated image.


Featured Video For You
This cyborg houseplant can automatically drive itself toward light

TopicsArtificial Intelligence

1.本站遵循行业规范,任何转载的稿件都会明确标注作者和来源;2.本站的原创文章,请转载时务必注明文章作者和来源,不尊重原创的行为我们将追究责任;3.作者投稿可能会经我们编辑修改或补充。

相关文章
  • Nate Parker is finally thinking about the woman who accused him of rape

    Nate Parker is finally thinking about the woman who accused him of rape

    2025-04-04 12:45

  • 青島隊不敵全華班申花仍小組墊底 掉入升降級附加賽漩渦

    青島隊不敵全華班申花仍小組墊底 掉入升降級附加賽漩渦

    2025-04-04 11:15

  • 李金羽	:多休一天很關鍵 再戰滄州是一個全新的挑戰

    李金羽 :多休一天很關鍵 再戰滄州是一個全新的挑戰

    2025-04-04 10:33

  • 儒尼奧爾:能進球要感謝球隊醫療組 他們幫我更快複出

    儒尼奧爾 :能進球要感謝球隊醫療組 他們幫我更快複出

    2025-04-04 10:12

网友点评