what's the tpr when far is at 0.0001%?

[hustzeyu]

what's the tpr when far is at 0.0001%?

[zhanglaplace]

@hustzeyu 这个没有测，自己只是随便跑的一个。抱歉

[piaopiaoxiaoye]

Can I use this caffemodel on Android platform or linux platform? Is this your already trained "mobilefacenet" model? Thank you! best regards!

[shiyuanyin]

@hustzeyu 这个没有测，自己只是随便跑的一个。抱歉

作者您好， 我用的vggface2,下载的训练集8631，的数据进行训练，损失函数和您的模型也是一样的，训练的时候从12左右的损失值开始减少，但是，并没有收敛，损失值在10左右震荡，不继续减小，作者您遇到过这个情况吗？？？可能有哪些原因昵

[zhanglaplace]

@piaopiaoxiaoye Andrioid : using ncnn and convert caffe model to ncnn model linux : Just compile caffe

[zhanglaplace]

@shiyuanyin learning_rate and batch_size ?

[shiyuanyin]

@zhanglaplace lr 0.1 开始， batchsize=64

[shiyuanyin]

@shiyuanyin learning_rate and batch_size 需要，预训练sortmax 的model吗？我直接训练，除了数据不是9136（我的8136），不能收敛，一直震荡

@zhanglaplace Andrioid : using ncnn and convert caffe model to ncnn model linux : Just compile caffe

需要，预训练sortmax 的model吗？我直接训练，除了数据不是9136（我的8136），不能收敛，一直震荡

[zhanglaplace]

@shiyuanyin Try to increase the learning rate. In addition, make sure your training data and labels are aligned。

[borisgribkov]

@shiyuanyin Did you solve the problem? I see the same behavior with ResNet-18 with ArcLoss and VGG2 training set, I guess that alignment is important

[borisgribkov]

@shiyuanyin I guess I fixed the problem, in case of not aligned vgg2 dataset training is very sensitive to M parameter, try to change