why use Normalized_Softmax_Loss instead of Unified_Cross_Entropy_Loss?

[TranThanh96]

Thank you for your great work.
I have read both your papers: Uniface and UniTSFace.
As I understand, in UniTSFace, you are combining sample-to-class loss with sample-to-sample-loss. In "sample-to-class loss", you are using Normalized_Softmax_Loss instead of Unified_Cross_Entropy_Loss (in paper Uniface). Can you explain why?