SAM produces a precise pixel mask for whatever you indicate with a click or box, without per-class training. SAM 2 adds a memory so a mask follows the object across video frames, which is the basis for rotoscoping and region-of-interest pipelines.
Definition
Segment Anything Model — a promptable segmenter that masks any object you point to or box, with SAM 2 extending this to track masks through video.
SAM produces a precise pixel mask for whatever you indicate with a click or box, without per-class training. SAM 2 adds a memory so a mask follows the object across video frames, which is the basis for rotoscoping and region-of-interest pipelines.
Also known as
Segment Anything, SAM 2