Meta SAM 3D

Segment Anything Model 3D (SAM 3D) is a computer vision model from Meta designed to reconstruct 3D information from single 2D images. Building on the Segment Anything ecosystem, SAM 3D predicts dense depth maps and per-pixel surface normals with a focus on generalization across diverse object categories and scenes. It is intended primarily as a research model to advance 3D understanding, rather than a turnkey content creation tool.

Users can leverage SAM 3D to obtain geometry-aware representations useful for downstream tasks such as 3D reconstruction, novel-view synthesis, robotics perception, AR/VR scene understanding, and human or object pose analysis. The project page provides technical details, model capabilities, datasets used, and links to code or weights where available.