I understand that the way AP is calculated lowers the threshold of any confidence, and for each threshold, it is calculated and graphed. Then, what is the threshold that is lowered when calculating the AP through R-CNN? Is that the result of the SVM?
For example, if the result of class 1 is 3.8 and the result of class 2 is 4.5 when the result of classes are true through SVM, is it determined as a bounding box of class 2 below threshold 4.5, and is it ignored at a higher threshold?
I'd appreciate your help.
