problem in cal_gradient_penalty?

[DreamsofGg]

just the same issue as caogang/wgan-gp#47
how do I understand the following code in

DeblurGAN/models/losses.py

Line 158 in 2499a87

gradient_penalty = ((gradients.norm(2, dim=1) - 1) ** 2).mean() * self.LAMBDA

especially dim=1, it seems that the function calculate pixelwise gradient norm rather than the full image, which is not consistent to formulation in the paper.
looking forward to your reply, thanks!
@KupynOrest