Segmentation
Classifying every pixel in an image so you know exactly which region belongs to what object.
Think of it like
Coloring each pixel of a map by country instead of just drawing borders.
Example
Medical imaging: segment tumor pixels (red), healthy tissue (blue), background (gray) so surgeon sees exactly what to remove.
How it actually works
Pixel-level prediction; many times slower than classification or detection. Encoder-decoder architecture: encoder downsamples (learn features), decoder upsamples (restore resolution). Skip connections preserve fine detail. Requires pixel-level annotations, extremely expensive (weeks of manual labeling per image).
For product teams
Critical for robotics, medical imaging, autonomous driving—need precise region boundaries.
For engineers
Encoder-decoder network outputs logits per pixel; cross-entropy loss on each pixel; typically uses U-Net or transformer architecture.
Related
- Object Detection — Refinement of.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome