Confuse robust l1 like l2

[huynhhoanghuy]

Sorry if I wrong, did you mistake about l1 norm? Your "robust l1" look like l2 norm without eps.
What effect do you have from eps?

GCNDepth/mono/model/mono_fm/net.py

Line 51 in 9fa7781

def robust_l1(self, pred, target):

[ArminMasoumian]

Hi, no we used the l1 norm, we just square root it and to the power of two instead of normal absolute. In the l2 norm, the sum is inside the square root but not in our case. Here is the explanation in the equation. This will increase the accuracy. Also, the eps. is just added to prevent getting zero results from loss.