SAM3

Model

SAM3 (Segment Anything Model 3) is Meta's latest foundation segmentation model in the Segment Anything series. It extends upon SAM and SAM 2 with improved text-prompt segmentation capabilities, robust object detection, and video object tracking. SAM3 serves as a universal perception backbone for various downstream applications including robotic manipulation, AR/VR, video understanding, and annotation systems. Its zero-shot generalization across diverse visual domains makes it a critical infrastructure component in the embodied AI ecosystem, used by projects such as SPARK for object grounding and perception scoring.