mAP summarises a detector's quality across all classes and confidence thresholds into one number. It rewards finding the right objects (precision) and not missing them (recall) at a given box-overlap level. It is how you compare two detectors fairly.
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
mAP lets you compare detectors fairly, but a high score on a public benchmark says little about your camera angles. This volume covers mAP in its computer vision section: detection metrics, IoU thresholds, precision and recall, and evaluation on domain data before deployment. Written by Nikolay Sapunov, CEO at Fora Soft.
Does a detector's mAP on benchmarks hold up on your footage?
Send us labelled samples, and we will measure mAP for candidate detectors on your data and recommend the one that fits your accuracy target. 250+ video products shipped since 2005.
