Image segmentation labels an image pixel by pixel, tracing each object's precise outline instead of just drawing a box around it.
In plain words
Where object detection gives a rough box, segmentation produces a mask: the set of pixels belonging to each object. Instance segmentation separates each individual object, so two people get two masks. Models like Mask R-CNN add a mask step to a detector, and the Segment Anything Model is built to be prompted, so it can handle new kinds of images and tasks without extra training.
A quick example
Imagine a photo editor's 'select subject' tool that runs a segmentation model. It finds the pixels that make up the person in the foreground, so you can remove or blur the background cleanly around their hair and hands.
Where you'll meet it
- Photo and video editing tools.
- Medical imaging and self-driving cars, where exact shapes matter.
Related
Sources
- PaperSegment Anything, Kirillov et al. (arXiv) · read 29 Sept 2026
- PaperMask R-CNN, He et al. (arXiv) · read 29 Sept 2026
- DocsInstance Segmentation, Ultralytics documentation · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.