SAM 2 is trained on the SA-V dataset that contains 51,000 real-world videos and more than 600,000 masklets. This dataset also consists of annotations for whole and partial objects to overcome challenges such as object occlusion, disappearance, or reappearance.
On July 29, 2024, Meta announced the release of its new AI-powered Segment Anything Model 2 (SAM 2) for object segmentation in images and videos. Backed by the success of its predecessor, SAM, which was designed for image segmentation, SAM 2 can detect and segment objects in images and videos.
To know more about Meta’s SAM 2, read here.
Object segmentation is a computer vision technique that separates images and video frames into distinct groups of pixels or segments to identify objects. It is most commonly used for image processing in self-driving vehicles, remote sensing, medical imaging, and document scanning.
Released under the Apache 2.0 license, users can prompt SAM 2 to segment any object in images and videos, including those it has not seen previously. It has been trained on the SA-V dataset and contains 51,000 real-world videos and more than 600,000 masklets. The ability to track fast-moving and dynamic objects makes SAM 2 suitable for object segmentation in videos.
Read More: Safeguarding Digital Spaces: The Imperative of Image Moderation
Meta announced that SAM 2 will revolutionize image and video-based content creation as it will simplify editing by automating segmentation using artificial intelligence. It is six times faster than its predecessor and will give users a better immersive experience in augmented reality (AR) and virtual reality (VR) applications.
Keeping up with its vision of open-source AI, Meta has open-sourced SAM 2 and the SA-V dataset on which the model was trained.
SAM was first introduced in 2023 as an AI model for image object segmentation. It was trained on the SA-1B dataset, which contains 1.1 billion segmentation masks collected from nearly 11 million licensed and secure images.
Since its launch, SAM has become highly popular as a segmentation tool in content creation, medicine, marine sciences, and satellite imagery. The success of SAM motivated Meta to unveil its upgraded version.
The AI landscape is advancing rapidly, and the release of SAM 2 will provide a much-needed push toward developing more efficient media processing tools. Meta’s vision of open-source AI has further raised the expectations of having easier access to more sophisticated AI solutions in the future.