I noticed that loc_loss uses l1 loss instead of smooth l1 loss. Does l1 loss perform better?

[TCBocean]

As the title says, does the l1 loss have a better performance than the smooth l1 loss, is there any relevant data? Because in your paper, the regression loss uses a smooth l1 loss.
I am looking forward to someone who can answer me, thank you.

[chun-chun123]

https://www.jianshu.com/p/19483787fa24

[TCBocean]

@chun-chun123
Yes, as the article says, smoothL1 has many advantages.

pysot/pysot/models/loss.py

Line 31 in 052b967

def weight_l1_loss(pred_loc, label_loc, loss_weight):

But I saw in the code that L1 loss was used, so I asked this question.