mAP
Evaluates the performance of your neural net for object recognition
... the same label and an IoU >= 0.5 (Intersection over Union greater than 50%). This "match" is considered a true positive if that ground-truth object has not been already used (to avoid multiple detections of the same object).