About L2 & TV loss, mean value

[mountains-high]

Good day,
Thank you for the nice job and for making it open-source ~
I have some doubts about these two losses. Could you please explain why you have taken the "mean of L2 loss and TV loss"?

DIODE/deepinversion_yolo.py

Line 178 in 80a396d

l2_loss = torch.norm(input[0], dim=1, p=2).mean()

and

DIODE/deepinversion_yolo.py

Line 245 in 80a396d

loss_var_l1 = (diff1.abs() / 255.0).mean() + (diff2.abs() / 255.0).mean() + (

Thank you ~

[mountains-high]

These two below gotten from the original source DI/cifar10:
https://github.com/NVlabs/DeepInversion/blob/6d64b65c573a8229844c746d77993b2c0431a3e5/cifar10/deepinversion_cifar10.py#L184
loss_var = torch.norm(diff1) + torch.norm(diff2) + torch.norm(diff3) + torch.norm(diff4)

and

https://github.com/NVlabs/DeepInversion/blob/6d64b65c573a8229844c746d77993b2c0431a3e5/cifar10/deepinversion_cifar10.py#L177
loss = loss + l2_coeff * torch.norm(inputs_jit, 2)
Which will be correct one? Thanks