pretrained models

[kiyse0713]

Could you please upload your highest accuracy pretrained models? @qidiso

[qidiso]

because the training is not end, if i finished ,i will

[qidiso]

you can visit my log file:
https://github.com/qidiso/mobilefacenet-V2/blob/master/wd1e-6-0.01.log

[qidiso]

I suggest you take second steps to train with s=128, and the third step is to train with s=64, which will make the value of LFW higher.The second step is not model-y1-arcface, 80，you can take the model-y1-sofmax,80 instead. i training order is : softmax,s=64 ,s=128,s=64.this make the lfw higher.

[chenkingwen]

what's the hyper-parameters of the first step by using softmax to train the net reaching the accuracy 99.2% on lfw? thank you very much.@qidiso

[ramanathan831]

@qidiso Can you provide the exact commands of each steps in the ReadMe file please?