Why not apply latest face recognition loss function?

penincillin / DREAM

This is the public repository for our accepted CVPR 2018 paper "Pose-Robust Face Recognition via Deep Residual Equivariant Mapping"

http://mmlab.ie.cuhk.edu.hk/projects/DREAM/

BSD 2-Clause "Simplified" License

388 stars 98 forks source link

Why not apply latest face recognition loss function? #27

Closed Joker316701882 closed 5 years ago

Joker316701882 commented 6 years ago

@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. For example, cosface loss or arcloss. Have you tried that?

(In latest face recognition loss function, features and weights of last fc both need to be normalized, is it because normalized feature will hurt the performance in your pose robust algorithm?)

Thank you : )

penincillin commented 5 years ago

Thanks for asking, the DREAM block is compatible with any face recognition loss, you could try them.