評價指標收集

TP/TN/FP/FN的定義

TP: 預測爲1(Positive)，實際也爲1(Truth-預測對了)

TN: 預測爲0(Negative)，實際也爲0(Truth-預測對了)

FP: 預測爲1(Positive)，實際爲0(False-預測錯了)

FN: 預測爲0(Negative)，實際爲1(False-預測錯了)

Accuracy/Precision/Recall/F1的定義

Accuracy = (預測正確的樣本數)/(總樣本數)=(TP+TN)/(TP+TN+FP+FN)

Precision = (預測爲1且正確預測的樣本數)/(全部預測爲1的樣本數) = TP/(TP+FP)

Recall = (預測爲1且正確預測的樣本數)/(全部真實狀況爲1的樣本數) = TP/(TP+FN)

F1 = 2*(Precision*Recall)/(Precision+Recall)

IOU的定義

IoU = TP/(TP+TN+FP)