Segment Anything Model 2 (SAM 2)
Real-time promptable segmentation for images and video, ~44 fps, with SA-V dataset of 51K videos and 600K+ masklets; Apache 2.0.
Added a streaming memory mechanism so one model segments and tracks objects across video frames; video annotation was 8.4x faster than per-frame manual labelling. SAM 2.1 update followed 2024-09-30.
- Date
- Monday, 29 July 2024
- Lab
- Meta
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Inference speed | ~44 fps 6x faster than SAM on images; 3x fewer interactions in video | company |
Sources
This record was checked and corrected against its sources on 6 October 2026. How we check