Detectors often fire several boxes on one object. Non-max suppression (NMS) keeps the highest-confidence box and discards others that overlap it beyond an IoU threshold. Newer detectors aim to be NMS-free to save this post-processing step.
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
Non-max suppression decides which of several overlapping boxes survives, and a bad threshold either duplicates objects or merges people standing close together. This volume covers non-max suppression in its computer vision section: detection and bounding boxes, IoU thresholds, NMS-free detectors, and tracking. Written by Nikolay Sapunov, CEO at Fora Soft.
Seeing duplicate or merged boxes after non-max suppression?
Send us frames where detection goes wrong, and we will tune non-max suppression or recommend an NMS-free detector with measured accuracy. 250+ video products shipped since 2005.
