Intersection over Union (IoU)

Intersection over Union, or IoU, is a metric for comparing two regions. It divides the area shared by a prediction and its ground truth by the total area covered by either region. A value of one means perfect overlap, while zero means no overlap. Computer vision systems use IoU to evaluate object-detection boxes and segmentation masks, often applying a threshold to decide whether a detection counts as correct. Different thresholds can substantially change reported performance, especially for small objects. IoU can also be averaged across classes, images, or instances, so comparisons require the same aggregation rules and treatment of empty or ignored regions.