Demo examples present how SAM 2 tracks and isolates video parts | Photograph credit score: Screenshots obtained from Meta and compiled in Canva
Meta has unveiled a brand new AI mannequin referred to as Phase Something Mannequin 2, or SAM 2, which it claims can decide which pixels belong to a specific object in movies.
The Phase Something mannequin, launched final yr by Fb’s dad or mum firm, helped develop options on Instagram corresponding to “Backdrop” and “Cutouts.” SAM 2 is meant for video media, and Meta claims that SAM 2 may “phase any object in a picture or video, and observe it constantly throughout all frames of a video in actual time.”
Nonetheless, along with social media and combined actuality use circumstances, Meta defined that its older segmentation mannequin was utilized in ocean analysis and catastrophe reduction, in addition to most cancers detection.
(Uncover the complexities of our digital world on The Interface podcast, the place enterprise leaders and scientists share insights that form tomorrow’s innovation. The Interface can be accessible on YouTube, Apple Podcasts, and Spotify.)
How segmentation works in SAM 2 by Meta | Photograph credit score: Meta
“SAM 2 may be used to trace a goal object in a video and assist extra rapidly annotate visible knowledge to coach machine imaginative and prescient techniques, together with these utilized in autonomous automobiles. It may additionally allow inventive methods to pick and work together with objects in real-time or in dwell video,” Meta stated in a weblog put up.
The social media firm has invited customers to check the mannequin, which will likely be launched underneath a “permissive” Apache 2.0 license.
Meta CEO Mark Zuckerberg mentioned the brand new mannequin with Nvidia CEO Jensen Huang, praising its scientific functions, TechCrunch reported.
