About recognition

Hi：
I'm confused about two question ，could explain for me.

first i want to know the data you train the net like deepid is using single patch？

the next is have you take a classification to realize the Face recognition with the Features extract from the net like SVM，how much is the accuracy for that。