Segment Anything Model (SAM)
Definition
Delineate objects in images based on cues.
Meta's Segment Anything Model generates segmentation masks for image areas, for example based on points or outlines as input.
Simply explained
A marked object is outlined as precisely as possible in the image.
Why it matters in practice
Segmentation can support image editing and the preparation of visual datasets.
Example from business
A hypothetical team marks products in images and checks the generated masks.
Limits and pitfalls
- The mask does not have to be correct; segmentation does not automatically mean that the object type was recognized.
Origin & history
Meta published the original SAM project in 2023.
Use in mid-sized companies
- Check masks on representative images, especially with fine edges and overlaps.
Frequently asked questions
Is SAM the same as object recognition?
No, the focus is on spatial delineation.
Sources
Editorially reviewed on: