There is a problem with the understanding of training classifier A first and then training classifier B

[Mr-Da-Yang]

Hello！
for what you described in the article, first train classifier A, then train classifier B, but in vgg_v1.py, add these two losses, and then the main function reverses this loss. Shouldn't the losses of classifer A,B be passed back separately in the main function?
Thanks for the guidance!

[xiaomengyc]

Hi，
We apply two classifiers. The final loss value is computed by adding the cross-entropy loss of them. We do not apply any reverse functions.