Decoder. plain-English AI glossary

Instance Segmentation

● Core

Identifying each distinct object and outlining its pixels - distinguishing "dog 1," "dog 2," etc., not just "dog."

Think of it like

Semantic segmentation colors all dogs red; instance segmentation outlines each dog separately.

Example

A photo of three dogs: semantic segmentation marks all dog pixels. Instance segmentation outlines each dog separately (dog #1, dog #2, dog #3).

How it actually works

Combines object detection (bounding box per object) and semantic segmentation (per-pixel labeling). Typically two-stage: Faster R-CNN for boxes, then mask prediction in each box (Mask R-CNN). One-stage alternatives: YOLACT, CondInst. Challenges: dense predictions on high-res images (slow), overlapping objects (which pixels belong to which?), rare categories.

For product teams

Autonomous vehicles, medical imaging (individual organ segmentation), robot manipulation.

For engineers

Mask R-CNN: Faster R-CNN + mask branch. Datasets: COCO (80 classes). Metric: Average Precision at IoU=0.5 and 0.75, Average Recall.

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