Segment Anything is a brand-new AI model released by Meta AI, which can "cut" any object in any image with one click. Segment any Model(SAM) is an impressionable segmentation system, which can be extended to unfamiliar objects and images without extra training.

This model will bring great influence to the research in the field of computer vision, because it can solve many practical problems, such as scientific image analysis, image editing, three-dimensional reconstruction and so on.

It can also be flexibly combined with other AI systems, such as using user's eye movements or object detectors to select objects to be segmented. It can also generate multiple effective masks to deal with vague hints.

The model is used as a "data engine" in iterative training, and the Segment Anything model (SAM) and its data are used to interactively mark pictures and update the model.

This iteration is repeated many times to improve the quality of the model and data set. More than11000000 pictures with rights and privacy protection have been collected, which is the largest segmentation data set at present (far exceeding the previous data set).