meta
On Wednesday, Meta unveiled an AI model called the Segment Anything Model (SAM). The model can identify individual objects in images and videos, even those that were not encountered during training, he reports Reuters.
According to Meta’s blog post, SAM is an image segmentation model that can isolate specific objects in an image in response to text prompts or user clicks. Image segmentation is a computer vision process that involves dividing an image into multiple segments or regions, each representing a specific object or region of interest.
The purpose of image segmentation is to facilitate image analysis and processing. Meta also says the technology will help him understand the content of his web pages, augmented reality applications, image editing, and support scientific research by automatically locating and video-tracking animals and objects. thinking about.
Creating an accurate segmentation model typically “requires AI training infrastructure and highly specialized work by technical experts with access to large volumes of carefully annotated in-domain data,” he says. says Meta. By creating SAM, Meta hopes to “democratize” this process by reducing the need for specialized training and expertise, and encourage further research into his vision of computers.
In addition to SAM, Meta created a dataset called “SA-1B”. It contains 11 million images licensed from “big photo companies” and 1.1 billion segmentation masks generated by its segmentation model. Meta makes SAM and its datasets available for research purposes under the Apache 2.0 license.
The code (unweighted) is now available on GitHub, and Meta has created a free interactive demo of its segmentation technology on a special website. Using the demo, visitors can upload a photo and click “Hover & Click” (select an object with the mouse), “Box” (select an object in a selection box), or “Everything” (all objects within an object). which tries to automatically identify the object). image).
Benj Edwards / Meta
Image segmentation technology is not new, but SAM is notable for its ability to identify objects not present in the training dataset and for its partially open approach. Also, the release of the SA-1B model could spark a new generation of computer vision applications, just as his LLaMA language model for Meta has already inspired derivative projects.
According to Reuters, Meta CEO Mark Zuckerberg 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 used SAM-like technology within Facebook to tag photos, moderate content, and improve user interaction on Facebook and Instagram. I was making recommendations post decisions.
Meta’s announcement comes amid a fierce race for big tech companies to dominate the AI space. His ChatGPT language model for Microsoft-backed OpenAI will garner widespread attention in the fall of 2022, triggering a wave of investment that could define the next major business trend in technology beyond social media and smartphones. I was.
