Meta, which is owned by Facebook, announced the release of an artificial intelligence (AI) model on Wednesday. The model, called the Segment Anything Model (SAM), can recognize individual objects present within an image or video, even if it has not been trained on those particular items before.
Meta also shared a dataset of image annotations that it claimed was the largest one ever of its kind. Using SAM, objects can be identified through text prompts or by clicking on them.
In one demonstration, typing the word "cat" resulted in SAM creating boxes around each cat present in a photo. Meta plans to use generative AI, reminiscent of Microsoft-backed ChatGPT, to create new content and features for its apps.
However, the company has not yet released such a product. Currently, Meta employs similar technology to SAM for activities like moderating prohibited content on Facebook and Instagram, tagging photos, and recommending posts to users.
SAM's model and dataset will be available for download under a non-commercial license, and users must agree to only use the model for research purposes when uploading their own images to an accompanying prototype.
Source: https://www.reuters.com/technology/...-can-identify-items-within-images-2023-04-05/
Meta also shared a dataset of image annotations that it claimed was the largest one ever of its kind. Using SAM, objects can be identified through text prompts or by clicking on them.
In one demonstration, typing the word "cat" resulted in SAM creating boxes around each cat present in a photo. Meta plans to use generative AI, reminiscent of Microsoft-backed ChatGPT, to create new content and features for its apps.
However, the company has not yet released such a product. Currently, Meta employs similar technology to SAM for activities like moderating prohibited content on Facebook and Instagram, tagging photos, and recommending posts to users.
SAM's model and dataset will be available for download under a non-commercial license, and users must agree to only use the model for research purposes when uploading their own images to an accompanying prototype.
Source: https://www.reuters.com/technology/...-can-identify-items-within-images-2023-04-05/