Meta unveils SAM, a new AI model that can spot and segment items inside an image

Facebook's parent company Meta unveiled its latest artificial intelligence (AI) tool SAM, an abbreviation for Segment Anything Model. The futuristic model, like its name suggests, possesses the ability to spot and "segment" various items in an image or video. In a paper released on Wednesday, the tech giant said that SAM "is designed and trained to be promptable, so it can transfer zero-shot 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," Meta explained on its official blog. According to the company's research department, the Segment Anything Model can spot any object inside an image or a clip, even if it has never witnessed some of those items during its training stage. 

The new AI model works rather simply, and allows users to select objects by either tapping on them or writing them in text prompts. A demo of the process showed that writing "cat" alerted the tool to create boxes around cats in one image. SAM's release comes alongside a dataset that will "foster research into foundation models for computer vision".  

SAM to broaden access to existing technology 

While the AI tool is brand-new, Meta has used similar technology in the past to carry out basic activities such as moderating inappropriate content, tagging images, and deciding which posts can be sent as recommendations for users on Facebook and Instagram.

However, the company believes that the release of SAM would widen the scope and the extent of such type of technology. The new model and its corresponding dataset will soon be up for downloading under a non-commercial license. To make sure that the technology is not used for malicious purposes, users will have to agree to upload their pictures for research-based reasons only.