Facebook owner Meta on Wednesday unveiled an artificial intelligence model that can select individual objects from within an image, along with a dataset of image annotations it said was the largest ever of its kind.
The company’s research department said in a blog post that its Segment Anything Model (SAM) can identify objects in images and videos, even if those items were not encountered in training.
SAM allows you to select objects by clicking on them or by entering a text prompt. In one demonstration, I wrote the word “cat” and the tool now draws boxes around each of the multiple cats in the picture.
Ever since Microsoft-backed OpenAI’s ChatGPT chatbot caused a sensation in the fall, sparking a wave of investment and a race to dominate the space, tech giants have been touting breakthroughs in artificial intelligence.
Meta hinted at several features that deploy the type of generative AI popularized by ChatGPT. ChatGPT creates entirely new content instead of simply identifying or classifying data like other AIs, but has yet to release a product.
Examples include tools that spin up surrealist videos from text prompts and tools that generate children’s book illustrations from prose.
CEO Mark Zuckerberg has said that incorporating such generative AI “creative aids” into Meta’s apps is a priority this year.
Meta already uses SAM-like technology internally for activities such as tagging photos, moderating banned content, and deciding which posts to recommend to users on Facebook and Instagram.
The company says the release of SAM will broaden access to this type of technology.
SAM models and datasets are available for download under a non-commercial license. Likewise, users who upload their own images to the accompanying prototypes must agree to use them for research purposes only.