Meta introduced an AI model that can detect objects within pictures and videos even if they were not part of the training set. Segment Anything Model (SAM) will highlight all the felines in a given photo by just typing a word. 

RUSSIA-US-INTERNET-META
(Photo : KIRILL KUDRYAVTSEV/AFP via Getty Images)
Photo taken on October 28, 2022, shows the META logo on a smartphone screen in Moscow.

Launching Segment Anything

AI models are usually trained on existing materials to detect objects. However, Meta is launching its model that spots items without help. The Segment Anything AI model can identify an object within pictures even if they were not part of the training set. Users can select items by clicking them or using free-form text prompts.

Aside from this, the model can also work in conjunction with other AI models and helps to reconstruct an object using 3D in a single image or draw from views through a mixed reality headset. As per Engadget's report, Segment Anything can lessen the need for AI's additional training.

Meta's Research Division stated that this is the largest segmentation data set to date, with over 1 billion masks on 11M licensed and privacy-respecting photos. SAM was built and trained to be prompt for the model to be able to transfer zero-shoot to new image distributions and tasks. 

"We evaluate its capabilities on numerous tasks and find that its zero-shot performance is impressive - often competitive with or even superior to prior fully supervised results," the company stated in its blog post. 

Segment Anything Model and its data set will be downloadable with a non-commercial license. Although creators can not use it for products as this model is intended for research and expanding access to the technology. 

Also Read: Meta's AI is Expanding as Mark Zuckerberg Forms New Team for Social Media Platforms

Gizmodo reported that SAM does a solid job of highlighting different objects in a photo as it is capable of outputting several masks despite the ambiguity about the object. This will be useful for image segmentation, both interactive and automatic.

Meta AI

As of the moment, Meta is using similar technology for the company to moderate banned content, recommended posts, and tag photos. But developers acknowledge how this existing technology is inaccurate at detecting the boundaries of some of its models as it might miss finer details. 

Reuters reported that Meta has been teasing to launch of several features that have the type of generative AI, popularized by ChatGPT. Through this technology, it will create brand-new content instead of identifying or categorizing data. 

Chief Executive Officer Mark Zuckerberg stated that incorporating generative AI into its applications is their utmost priority for this year. Meta is also not new to sharing AI technology with its translator for unwritten languages. 

Meta also plans to invest in AI at the end of 2023 to improve its ad creation in its application. The company already created a new team a couple of months ago, along with Chief Technology Officer Andrew Bosworth, who said that he and Zuckerberg are both spending the most time in.

Related Article: Meta AI Researchers Reveal Cicero, First AI to Rival Humans in Diplomacy

Written by Inno Flores

ⓒ 2024 TECHTIMES.com All rights reserved. Do not reproduce without permission.
Join the Discussion