Segment Anything(SAM)
SAM is Meta AI's open-source image segmentation model, producing high-quality masks from point or box prompts and segmenting entire images automatically. Trained on 11M images and 1.1B masks, it delivers strong zero-shot performance.
Human verified · · Submit a correction
Ideal for researchers, AI engineers, and labeling teams needing general-purpose segmentation; not for non-technical users expecting one-click, semantically labeled results.
Verified facts
What is Segment Anything(SAM)
SAM is Meta AI's open-source image segmentation model, producing high-quality masks from point or box prompts and segmenting entire images automatically. Trained on 11M images and 1.1B masks, it delivers strong zero-shot performance.
Key features of Segment Anything(SAM)
- Generating precise object masks from point or box prompts
- Segmenting every object in an image to build datasets
- Auto-drafting initial masks to cut labeling costs
- Embedding segmentation into cutout and image-editing products
Good for
- Strong zero-shot performance — no retraining for new domains
- Intuitive prompt-based interaction plus full-image auto mode
- Fully open-sourced by Meta with paper, demo, and code
Watch out
- Outputs masks without semantic labels; other models must supply meaning
- Raw model targets engineering teams; casual users rely on demos or third-party tools
- Highly specialized imagery may still require fine-tuning
How to use Segment Anything(SAM)
- Open the online demo at segment-anything.com
- Upload an image and prompt with points or boxes
- Switch to automatic mode to segment everything
- Deploy locally with code and weights from GitHub
- Read the paper, then fine-tune or wrap as a service
Who Segment Anything(SAM) is for
Difficulty: Advanced
- Generating precise object masks from point or box prompts
- Segmenting every object in an image to build datasets
- Auto-drafting initial masks to cut labeling costs
- Embedding segmentation into cutout and image-editing products
FAQ
Can SAM identify object categories?
SAM focuses on producing object masks (where things are) and does not output class labels (what they are); semantic recognition requires pairing with classification or detection models.
Is SAM free to use?
Yes. SAM is an open-source Meta project — model, code, and demo are free; costs come mainly from your own deployment compute.
Can non-programmers use SAM?
Casual users can try point-and-box segmentation in the official online demo; integrating the model into products requires engineering skills.
Sources and verification
Sources: segment-anything.com (opens in a new tab)
Verified: · Submit a correction →