It can not converge on non-pretrained model

[YihengJiang]

Dear Author, Thank you for your contribution. Here I have run your code, and I found that the model only can converge on pretrained pattern with triplet loss. When I run program with non-pretrained model, the triplet loss can not converged, and the value will stop to the margin,for example, if I set margin=0.5, the last loss after training is 0.5, too. So here I can not reproduce your result in your paper. is it real right?or something I am wrong, hope your response. Thank you

[YantaoShen]

Hi, could you tell me which paper are you reproducing?