SAM 3 and SAM 3D
SAM 3 segments and tracks every instance matching a text or exemplar concept; SAM 3D reconstructs objects and human bodies from one image.
Adds promptable concept segmentation with an open-vocabulary detector and a presence head, trained on 4M unique concept labels; paper claims 2x the accuracy of prior systems on the new SA-Co benchmark. SAM 3D Objects/Body extend the line to 3D.
- Date
- Wednesday, 19 November 2025
- Lab
- Meta
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Concept segmentation accuracy vs existing systems | 2x paper claim on SA-Co (images and video) | company |
Used in Meta's Edits app and Facebook Marketplace 'View in Room'.
Sources
- about.fb.com/news/2025/11/new-sam-models-detect-objects-create-3d-reconstructions/
- arxiv.org/abs/2511.16719
This record was checked against its sources on 6 October 2026. How we check