This method is proposed in paper: TypicFace: Dynamic Margin Cosine Loss for Deep Face Recognition. It improved AM-Softmax with a hyperparameter.
From the experiment result, we can find:
m = 0.3 d=0.875 will get a better result.
This method is proposed in paper: TypicFace: Dynamic Margin Cosine Loss for Deep Face Recognition. It improved AM-Softmax with a hyperparameter.
From the experiment result, we can find:
m = 0.3 d=0.875 will get a better result.