๐ง SAM 3
A unified promptable concept-segmentation model that detects, segments and tracks objects in images and videos. It handles open-vocabulary text concepts and visual prompts.
At a glanceโ
| Field | Details |
|---|---|
| Model type | Concept-prompt image and video segmentation |
| Research team / organization | Meta AI Research |
| Input / modality | 2D images and videos; not a native 3D-volume segmenter. |
| Prompt / interaction | Short text concepts, image exemplars, points, boxes and masks; video tracking and refinement are supported. |
| Code license | Code and checkpoints use Meta's SAM License, not an unrestricted permissive license. Review the official terms before use. |
๐ Paper and publicationโ
SAM 3: Segment Anything with Concepts โ arXiv:2511.16719 (2025 technical report/preprint).
๐ฆ Checkpoints and accessโ
The official Hugging Face checkpoint is gated: request access, accept its conditions and authenticate with hf auth login. The repository currently also documents SAM 3.1 checkpoints; pin the exact checkpoint/repository revision when reproducing SAM 3.
The code, model checkpoint and any required foundation-model checkpoint can have different terms. Review each linked source before use or redistribution.
โ๏ธ Official setup and inferenceโ
conda create -n sam3 python=3.12
conda activate sam3
# Install a PyTorch/CUDA combination supported by the current upstream README:
pip install torch==2.10.0 torchvision --index-url https://download.pytorch.org/whl/cu128
git clone https://github.com/facebookresearch/sam3.git
cd sam3
pip install -e .
hf auth login # after checkpoint access is approved
For inference, follow the official image predictor notebook or the build_sam3_image_model() + Sam3Processor example in the README.
โ ๏ธ Version and practical notesโ
SAM 3 natively handles 2D images and videos. Any 3D medical workflow adds an adaptation and a slice/clip policy that should be documented.
๐ Licenseโ
Code and checkpoints use Meta's SAM License, not an unrestricted permissive license. Review the official terms before use.
Citation: Cite the paper above and the upstream project as directed by its authors. This repository links to the original paper; it does not redistribute the code or weights.
Source pages checked: 27 October 2026. Upstream environments, weights and access policies may change; use the linked project as the source of truth.