no-saturated loss

[MengZhen-Chi]

in your paper you used no-saturated loss for D and G, could you told me the change of D loss in training process? In my training process, my D loss will down to 0.0139 after about 1k iters

[MengZhen-Chi]

my D loss code is :
D_adv = GANLoss(gan_type='wgan_softplus', real_label_val=1.0, fake_label_val=0.0, loss_weight=1.0)
L_adv_D = 0.01 * (D_adv(real_pred, True) + D_adv(fake_pred, False))
real_pred is D(ground_truth) , fake_pred is D(generated_sample)