Meta announced the Anything Model (SAM) segment, an AI model that identifies objects in images and videos even if they aren’t present in the training data set. SAM is an image segmentation model that responds to text prompts and clicks on objects in an image.

Image segmentation divides images into different segments based on the objects in the image. The technology facilitates the analysis and processing of images. Meta wants to use this technology to understand website content, augmented reality and image processing.

In addition, the technology can be used for scientific research, for example by automatically locating and tracking animals or objects. With SAM, Meta aims to make the creation of an accurate segmentation model more accessible to encourage further research into computer vision.

In addition to SAM, Meta has compiled a dataset called “SA-1B” that contains 11 million images. The company makes SAM and the dataset available for research purposes under an Apache 2.0 license.

Demo available

SAM’s code is currently available on GitHub. In addition, Meta offers a free interactive demo of its segmentation technology. In the demo, you can upload a photo and use tools like “Hover & Click” (select objects with the mouse), “Box” (select objects within a selection box), or “Everything” (where the technology automatically selects every object within it). try to recognize image.

Not new, but noticeable

While the image segmentation technology isn’t new, SAM certainly stands out. The reason for this is that the technology is able to identify objects that are not present in the training dataset.

In addition, the deployment of the SA-1B model could enable a new generation of computer vision applications. The same happened with Meta’s LLaMA language model, on which many side projects are already based.

Meta CEO Mark Zuckerberg confirmed earlier this year that he’s already working hard to bring generative AI into the company’s apps. While Meta has yet to release a commercial product using this type of AI, it is already working internally with a SAM-like tool to tag photos, moderate content, and determine recommended posts.