qidiso

Results 28 comments of qidiso

i make a mistake, sorry.i can download now!

train_softmax.py is from the insightface https://github.com/qidiso/insightface/blob/master/src/train_softmax.py

it's right.you need more epochs.

maybe you can try --margin-s 64 first!

may because your batch size is small!

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

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

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...

your batch size is too small...--wd is small too