Meta has released an artificial intelligence model, SAM, that can detect specific objects within an image, as well as the world’s largest dataset of image annotations.
In a blog post, the company’s research division stated that its Segment Anything Model, or SAM, could recognise objects in images and videos even when it had not encountered those items during training. Objects can be chosen using SAM by clicking on them or typing text prompts. Writing the word “cat” caused the tool to draw boxes around each of several cats in a picture in one demonstration.
Since Microsoft-backed OpenAI’s ChatGPT chatbot became a sensation, sparking a wave of investments and a race to control the space, big tech companies have been highlighting their artificial intelligence triumphs.
Although Meta has not yet released a product, it has teased several features that use the type of generative AI popularised by ChatGPT, which generates entirely new content rather than simply identifying or categorising data like other AI. A tool that produces surrealist videos from text prompts is one example, as is another that generates children’s book images from prose.
According to CEO Mark Zuckerberg, incorporating such generative AI creative aids into Meta’s apps is a goal this year. Internally, Meta already employs SAM-like technology for tasks such as tagging photos, moderating prohibited material, and determining which posts to suggest to Facebook and Instagram users.
According to the company, the release of SAM will increase access to that sort of technology. The SAM model and dataset will be made accessible for non-commercial use. Users who submit their own images to an accompanying prototype must also agree to only use it for research purposes.