how did you realize the adversarial process in your code?

[maliho0803]

hi, in you paper, in order to suppress the problem of order collapse, you introduce the adversarial loss. however, I did not find adversarial process, you just minimize the loss_div?

if I loss some information about your code?

[chenbinghui1]

@maliho0803 Hi, as in paper, the adv loss is minimized for adv net, and maximized for HOA, thus we use a gradients reverse layer to implement the maximizing process. And the code is in Line82 of losses.py