Facebook owner Meta releases AI model to detect items within images and is called SAM

Meta, the parent company of Facebook, unveiled on Wednesday a brand-new artificial intelligence model that is capable of locating objects within images. It is referred to as "segment anything" and aims to provide a more general understanding of images and videos as well as accelerate segmentation research.

The ability to determine which pixels in an image belong to a specific object is known as segmentation. Meta said its drive expects to "democratize" division, which is utilized in different applications, from analyzing scientific imagery to editing photographs.

To create an accurate segmentation model for particular tasks, technical experts with access to AI-trained infrastructure and large amounts of data typically need to perform highly specialized work. Meta's general Segment Anything Model (SAM) and Segment Anything 1-Billion mask dataset (SA-1B) have been made available as part of the new project.

Both plan to empower a wide arrangement of applications and encourage further research into establishment models for computer vision.

SAM can be used for a wide range of things, including underwater photography. It can improve creative applications like video editing and extracting parts from images for collages. Additionally, it could be utilized to enhance scientific investigations of natural occurrences on Earth or even in space.

Meta stated, "We are excited by the many potential use cases we haven't even imagined yet." "We believe the possibilities are broad." SAM and the SA-1B dataset are both available under a permissive open-license framework for research purposes.

Meta stated, "Our objective was to build a foundation model for image segmentation… a promotable model that is trained on a variety of data and can adapt to specific tasks, analogous to the use of prompting in models of natural language processing."

"However, in contrast to images, videos, and text, the segmentation data required to train such a model is not easily accessible online or anywhere else. "So, with Segment Anything, we set out to simultaneously create a segmentation dataset of unprecedented scale and develop a general, prompt segmentation model."

Meta claimed that the SA-1B dataset it released on Wednesday was the largest to date.

More than 1.1 billion segmentation masks from approximately 11 million licensed and privacy-preserving images are included in the final dataset.