Face verification
Face verification, one to one
Face identification / recognition
Face recognition, one-to-many
Closed set: This person is known to be in the data set
Kaiji: I don’t know if this person is in this data set
Face alarm ratio ( generally set to 1/10000 )
Number of wrong faces/total number of faces
Wrong face recognition: the number of two faces from different faces but recognized as the same face
Reference article: https://arxiv.org/pdf/1502.00873.pdf
In the test on lfw, the database can be divided into 10 parts, 9 parts of training, 1 part of test, 10 times in total, and 10 accuracy is obtained, so the final result will be similar to the form of 99.53+-0.1%