Open onlinehuazai opened 3 years ago
In the trainig progress, the trainig loss is decreasing but the accuracy on the training dataset is always zero. I don't know why this happens?
in arcface paper, the author set s=64,m=0.5, but i don't know it is correct to this project
the acc is 0, triplet loss and CE loss can obtain 0.5 acc, is hyper-parameters(s and m)?