IoU divides the overlap area of two boxes by their combined area. A detection counts as correct when its IoU with the ground-truth box clears a threshold, often 0.5. It also drives non-max suppression and tracker matching.
The book · Volume 6 of 7
AI for Video: A Complete Guide to Intelligence in Video Products: From Models and Inference to Computer Vision, Speech, Search, and Generative Video
IoU is the threshold that decides whether a detection counts as correct, so changing it shifts every accuracy number. This volume covers IoU in its computer vision section: detection and bounding boxes, evaluation metrics, non-max suppression, and matching in multi-object tracking. Written by Nikolay Sapunov, CEO at Fora Soft.
Unsure your detector's accuracy numbers mean what you think at your IoU threshold?
Send us your evaluation setup, and we will review IoU thresholds, metrics and test data and report how the model performs on your use case. 250+ video products shipped since 2005.
