Meta unveils AI model for object detection and largest dataset of its kind

According to a recent blog post by Meta’s research division, the company has developed an advanced object recognition model called the Segment Anything Model (SAM). SAM is designed to identify objects within images and videos, even if it has not encountered them during its training phase. The model allows users to select objects by clicking on them or using text prompts, such as the word “cat.” In a demonstration, SAM was able to accurately draw boxes around multiple cats in a photo in response to the written prompt.

Meta has been using technology similar to SAM internally for tasks such as tagging photos, moderating prohibited content, and recommending posts to Facebook and Instagram users. The company has stated that the release of SAM will expand access to this type of advanced technology beyond their internal operations.

The company has made the SAM model and dataset available for download under a non-commercial license. However, those who upload their own images to the accompanying prototype must agree to use the tool only for research purposes.

“In the future, SAM could be used to help power applications in numerous domains that require finding and segmenting any object in any image. For the AI research community and others, SAM could become a component in larger AI systems for more general multimodal understanding of the world, for example, understanding both the visual and text content of a webpage. In the AR/VR domain, SAM could enable selecting an object based on a user’s gaze and then “lifting” it into 3D,” said Meta in its blogpost. 

The technology giant has suggested that SAM could have several applications for content creators, including the ability to extract image regions for use in collages or video editing. Additionally, the model could prove useful in scientific research, allowing scientists to locate and track animals or objects of interest within video footage of natural occurrences on Earth or in space.