why there is only two softmax layer in prototxt

[ZhiweiYan-96]

hi~

I have read your paper, but I only find two softmaxloss layer is ' res_e1_train_val.prototxt' . But there are three softmaxloss Layers in paper. May be it is due to 'iterative training' ? Thanks for your help.

[hshustc]

In the iterative training, there are two losses involved. And in most cases, it is enough to set max_iter=1 to obtain the perform gain. Besides, the way to train with max_iter=2 is similar.

[ZhiweiYan-96]

So if the max_iter=2. I need three prototxt files which one of these is used in ite_0 and the other two are used in ite_2k and ite2kpulus1 ?

[ZhiweiYan-96]

The other problem is when three losslayer are needed? The visualization of DualNet has three losslayers in supplementary material. thxs

[hshustc]

1. In each training phase, only one prototxt is needed in my implementation.
2. Three loss layers are need in the jointly finetuning. The loss function is presented in the paper.

[ZhiweiYan-96]

thanks for your response! i will read it again.