SAM 2 is Meta's 2024 follow-up to Segment Anything that outlines and tracks objects through videos as well as in still images.
In plain words
The first SAM worked on single pictures. SAM 2 treats an image as a video with one frame, so the same model handles both. It uses a transformer with a streaming memory, letting it remember what an object looked like in earlier frames and follow it in real time. Meta trained it on SA-V, the largest video segmentation dataset at the time, and it needed about three times fewer user clicks than earlier video methods. Improved SAM 2.1 checkpoints followed in September 2024.
A quick example
In a video of a dog running through a park, click on the dog in the first frame. SAM 2 keeps a mask on the dog in every following frame, even as it moves and turns.
Where you'll meet it
- Following objects through video clips.
- Tracking objects in video for research.
Related
Sources
- PaperSAM 2: Segment Anything in Images and Videos, Ravi et al. (arXiv) · read 29 Sept 2026
- Docssam2, Meta (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.