Meta Launches an AI Tool That Can Segment Any Object in Any Image

Meta has launched an AI-based image segmentation tool called the "Segment Anything Model" (SAM), which can identify and separate any object in a picture. SAM is an interactive segmentation system that allows users to "cut out" any object in any image by clicking or drawing a box around it. This technology can be used for image editing, biological or security image analysis, or to quickly and efficiently recognize different parts of an image while wearing a VR headset.

Meta's tool is based on a huge dataset containing 11 million images and 1.1 billion segmentation masks. Meta claims this is "the largest segmentation dataset ever" and has made it open source for other researchers and developers to use. Meta hired 130 human annotators from Kenya to label the images in the dataset manually and automatically.

Meta showcased the tool to the public on Wednesday (April 5) and provided a browser-based demo version that users can try online for free. According to Meta, SAM can accurately identify and segment objects in images in complex and ambiguous situations, and can even interact with user input such as gaze, clicks or text.