No grad_fn for "self.loss_G_chamfer2"

[gerwang]

Hello,
Thanks for your great work! I'm pretty interested in your distance transform loss. But when I want to test the loss separately, I find there is no grad_fn for self.loss_G_chamfer2

APDrawingGAN/models/apdrawing_gan_model.py

Line 308 in 2b0173a

self.loss_G_chamfer2 = (dt1gt[(fake_B_gray<0)&(fake_B_gray_line1<0)].sum() + dt2gt[(fake_B_gray>=0)&(fake_B_gray_line2>=0)].sum()) / bs * self.opt.lambda_chamfer2

Since real_B_gray is input label and doesn't require grad, I believe self.loss_G_chamfer2 should have no grad_fn. So what is the purpose to calculate it? Or is it redundant?