USA: We've all had that one perfect photo that we couldn't post anywhere due to an obtrusive background. SAM, however, is Meta's solution to this issue.
Anything in an image can be extracted using SAM, or Segment Anything Model. According to the company, segmentation can be made simple using AI. Additionally, the Segment Anything 1-Billion mask dataset was made available (SA-1B).
AI's field of computer vision is rapidly developing. For a while now, businesses like Google and Amazon have been working on computer vision. Image segmentation is a crucial component of computer vision algorithms and technologies.
However, due to the complexity involved, most AI researchers are unable to create an accurate segmentation model for a given set of data. With SAM, Meta hopes to address that issue.
A generalised segmentation model is the SAM. It combines interactive segmentation and automatic segmentation, which are the two traditional methods of segmentation. Reduced dependence on "task-specific modelling expertise, training compute, and custom data annotation" is the goal of the model. SAM is a "promptable" model that is sufficiently all-encompassing to address a variety of tasks.
Also Read: Indian authorities alert Apple users to numerous security flaws
The dataset that SAM was trained on is what gives it its generality. Compared to other datasets, SA-1B has 400 times more segmentation masks (a particular area of an image separated from others) at over 1.1 billion. The size of the dataset allows SAM to generalise images and objects that were not used in its training.
By clicking on any object in the image or by "interactively clicking on points to include and exclude from the image," users can segment images with SAM, according to Meta.
Users can also draw a bounding box to instruct the model to segment an object. In cases of uncertainty, SAM can display multiple viable masks. The ability to use text prompts is not yet available.
Also Read: You might soon be able to "keep" disappearing messages on WhatsApp
Any field that calls for finding and segmenting objects will benefit from SAM. It will be simpler to use SAM with other more powerful AI systems thanks to its promptable design.
AR/VR is one of the use cases mentioned by Meta. SAM could lift into 3D and become a part of the user's gaze. Additionally, it will be helpful for research and content creation.