目标检测评价指标解释(precision, recall, mAP)
Reference:
https://nlp.stanford.edu/IR-book/html/htmledition/evaluation-of-ranked-retrieval-results-1.html
TP, FP, FN, TN
这些指标都源于下表
~ | ~ | 预测 | 预测 |
---|---|---|---|
~ | ~ | 1 | 0 |
实际 | 1 | TP | FN |
实际 | 0 | FP | TN |
Precistion Recall
Precision(预测正例正确占所有预测正例的比重) = TP / (TP + FP)
Recall = Ture Positive Rate(分类器识别的正例占) = TP/(TP + FN)
False positive rate = FP / (FP + TN)
False alarm = FP / (FP + TP)
2 / F_1 = 1 / Recall + 1 / Precision
mAP = \int_{1}^{1} P(r) dr
P是修正的Precision
R是Recall