Facebook owner Meta on Wednesday published an artificial intelligence model that can recognize objects within images, along with a data set of image annotations that it said was the largest of its kind ever.
The company’s research department said in a blog post that its model, known as the Segmentation of Anything Model, or SAM, can identify objects in images and videos even if he has not interacted with those elements during his training.
With SAM, items can be selected by clicking on them or typing in their name. In one demonstration, writing the word “cat” prompted the tool to place boxes on each of the cats in the image.
Big tech companies have been announcing their breakthroughs in artificial intelligence since Microsoft-backed OpenAI’s ChatGB chatbot made a splash, sparking a flurry of investment and a race for control.
META CEO Mark Zuckerberg said integrating such “creative AI assistants” into META applications is a priority this year.
Internally, Meta uses technology similar to the SAM model for tasks such as tagging photos, reviewing prohibited content, and determining which posts to filter for Facebook and Instagram users.
The company said the launch of the SAM model would expand access to this type of technology.