问题
Please explain how to calculate cumulative matching characteristic (cmc) curve? An example and some reading resources will be helpful?
回答1:
CMC curve is a precision curve that provides detection or recognition precision for each rank.
The horizontal line is rank of recognition, and the vertical line is precision percent. For example the first rank recognitions rate is indicated in first element, the second rank is indicated in second element and so on.
As a result a non-decreasing curve will result which indicates identification rate in each rank, and the identification rate for the last rank is equal to 1.
image of a sample CMC curve for identification rate per rank
回答2:
I would suggest going through the evaluation report for FRVT.
If you need some working code, you can use this implementation. Just keep in mind that it assumes your model labels are 1:K. For example, for 5 face classes, your models should be 1,2,3,4,5 and NOT 1,3,4,5,6.
回答3:
I have found this beautiful article The Relation Between the ROC Curve and the CMC explaining about CMC curve.
来源：https://stackoverflow.com/questions/47102074/cumulative-matching-characteristic-cmc