What is Instance Segmentation?
Computer VisionDeep LearningSegmentationYOLOIndustrial AI
What is Instance Segmentation?
Instance segmentation is the computer vision task of finding each object in an image separately and extracting each one's outline at pixel level.
Three Approaches Compared
| Task | Output | Weakness |
|---|---|---|
| Object detection | One box per object | The box includes pixels outside the object |
| Semantic segmentation | One class per pixel | Cannot separate two touching objects |
| Instance segmentation | A separate pixel mask per object | The most expensive to label |
When Do You Need It?
When you need to count objects and measure inside each one. For example, finding each tree's crown in a drone image and computing NDVI inside it. With boxes, the soil and grass between trees would spoil the measurement.
Common Models
- YOLOv8-seg / YOLO11-seg: fast, single-stage; close to real time in the field.
- Mask R-CNN: two-stage, slower, but strong on small objects.
- SAM (Segment Anything): produces masks zero-shot; often used to speed up labelling.
For a worked example see Walnut Crown Detection.