A Segment Anything Model of Meta can select individual objects within an image or video. SAM can find unidentified objects in its training.
Heaptalk, Jakarta — Facebook parent company, Meta, unveiled a new technology breakthrough in the Artificial Intelligence (AI) model, a Segment Anything Model (SAM), on Wednesday, (04/05). This technology can select individual objects from within an image or video and can identify cases where it had not encountered those items in its training.
“We released our general Segment Anything data set and our 1-billion mask (SA-1B) Segment Anything. This model is the largest-ever segmentation of data sets to enable a wide range of applications and encourage further research into the foundational models for computer vision,” said Meta representative through its official blog.
The company explained the launch of SAM would broaden access to this technology. The SAM model and dataset will be available for download under a non-commercial license. Users who submit their images to the accompanying prototype must also agree to operate them for research purposes only. Meta claimed the Segment Anything could also support a research study on natural and space events, such as detecting animals or objects to be tracked in videos.
AI technology has become the forefront of the national and global conversation in recent months after the release of the popular ChatGPT AI chatbot, which bolstered and sparked a race among tech giants to unveil similar tools.
Cited in Reuters, Meta‘s Chief Executive Officer, Mark Zuckerberg, has said that incorporating such generative AI “creative aids” into this Meta‘s applications is a priority this year.
As it is known, Meta company has already utilized a similar technology to SAM internally for several activities, such as tagging photos, moderating prohibited content, and determining which posts will be recommended to Facebook and Instagram users.
“The model is designed and trained to be prompt, enabling users to transfer zero-shot to new image distributions and tasks. We evaluate its capabilities on numerous tasks and find its zero-shot performance impressive – often competitive with or even superior to prior fully supervised results,” said Meta spokesperson.
