Facebook parent Meta introduced an AI model that can identify specific objects in videos and images, including objects it didn't learn during the training phase.
Meta says its Segment Anything Model (SAM)
aims to help democratize "segmentation," a task within computer vision
that involves identifying which pixels belong to which objects in an
image. Meta has launched a demo website for anyone to try out SAM.
- To
use SAM, users write text prompts or select objects within an image or
video. Typing the word cat, for example, causes SAM to draw a box around
each cat in photos.
- Meta is releasing SAM along with its
Segment Anything 1-Billion (SA-1B) mask dataset, which it says is the
largest segmentation dataset to date.
- The goals are to support new applications and research into computer vision foundation models, according to the firm.
- Meta
says it's opening up SAM and SA-1B to the research community. The model
is available under a permissive open license (Apache 2.0).
