yang0817manman
yang0817manman
Hi , Can you share your final-finetuned model of docface to me ?I want to test my dataset . Thanks a lot ! @seasonSH
@DuckJ Sorry,I don not have the author's model, either.
origin am-softman: s*(costheta - m) docface am-softmax: s*costheta - m so the parameter of "m" is different. @ronghuaiyang
thx, I sent e-mail according to this link that you recommended ,but i did not accept repy. @reshow
我利用“circle_loss.py”也是训练人脸，loss都没有收敛，一直处于震荡，请问这是什么原因？
请问你有没有试过人脸数据或者行人，类别数特别多，我用你分享的loss试验人脸，调试发现loss_p为0,无论怎么修改参数，打印的loss总是nan或0？ @qianjinhao