Intersection over Union (IoU)
Definition
Intersection over Union measures the overlap of two areas as the intersection divided by the union.
In object and segmentation evaluation, IoU is often calculated between predicted and actual bounding box or mask. The value range is between 0 and 1.
Simply explained
How to measure how much two cut-out areas overlap.
Why it matters in practice
Relevant for computer vision models and quality measurement in detection or segmentation.
Example from business
A model marks defects on component images; IoU compares the predicted defect mask with the verified reference.
Limits and pitfalls
- Consider only IoU without classification; choose thresholds arbitrarily; ignore small objects and class imbalance.
Origin & history
PASCAL VOC uses Intersection over Union to evaluate segmentations and box overlaps.
Use in mid-sized companies
- Possible applications: quality inspection, object detection, segmentation, and document layout analysis.
Frequently asked questions
Is IoU the general accuracy of correct classifications?
No, it evaluates an overlap. What does 1 mean? For non-empty union, complete overlap. For empty union, a special convention must be defined.
Sources
- The PASCAL Visual Object Classes Challenge · Everingham et al., The PASCAL Visual Object Classes Challenge
Sources checked on 10.10.2026.
Editorially reviewed on: