-
@penincillin
I noticed that in your source code, face recognition is treated as plain image classification task, without applying some latest loss function specifically designed for face recognition…
-
I try to reimplement the result by pytorch. However, I just get accuracy 96.7% on lfw when Arcloss is around 30. So I want to know how large arclose is when you reach 99%+ accuracy.
-
https://github.com/dmlc/gluon-cv/blob/master/gluoncv/model_zoo/se_resnet.py#L88