Meta has announced a new AI model called the Segment Anything Model (SAM). Unlike other AI models, SAM can identify individual objects in images and videos even if it hasn’t encountered them during training.
SAM is an image segmentation model that can isolate specific objects in an image in response to text prompts or user clicks. This is achieved by dividing an image into multiple segments or regions, each representing a particular object or region of interest. SAM is intended to facilitate image analysis and processing and could also be used in the content understanding of web pages, augmented reality applications, image editing, and scientific research.
Meta hopes to “democratize” the process of creating an accurate segmentation model by reducing the need for specialized training and expertise. To that end, Meta has built a dataset called SA-1B consisting of 11 million images licensed from “major photo companies” and 1.1 billion segmentation masks created with a segmentation model. Meta plans to make SAM and its dataset available under the Apache 2.0 license for research.
SAM is notable for its ability to identify objects not present in the training dataset and for its partially open approach. In addition to SAM, Meta has created a free interactive demo of their segmentation technique, allowing users to upload a photo and try to “Hover & Click”, “Box”, or “Everything” to select an object or identify all objects in the image.
Meta CEO Mark Zuckerberg has emphasized the importance of incorporating generative AI into the company’s apps this year. While Meta has yet to release a commercial product using this kind of AI, it has previously leveraged SAM-like technology internally with Facebook for photo tagging, content moderation, and recommendations on Facebook and Instagram.
The launch of SAM comes at a time of intense competition among big tech companies to dominate the AI ​​space. Microsoft-backed OpenAI’s ChatGPT language model received widespread attention in the fall of 2022, sparking a wave of investment that may define the next big business trend in technology beyond social media and smartphones.
