Work out map (mean average precision) instantly with clear inputs, formula shown and shareable results.
Average precision is the area under one class's precision-recall curve; mean average precision is the unweighted mean of those per-class values. Because the mean is unweighted, a rare class with poor AP drags the score down as much as a common one, which is exactly what makes mAP a useful test of balanced detection quality.
Mean average precision
mAP = (1/C) x sum of AP for each class c
Every class contributes equally regardless of instance count, so a class with 20 examples and AP 0.2 pulls the mean down as hard as a class with 20,000 examples.
State the threshold explicitly. mAP@0.5 is far more forgiving than mAP@[.5:.95] and the two are not comparable.