Why you take the abs() and rise to 2 in L_{s}? about rtfm HOT 3 CLOSED

Hello!

In Equation 6, as shown bellow, you calcule the L_{s} by:

However, in train.py, you calcule loss_abn and loss_nor by doing:

loss_abn = torch.abs(self.margin - torch.norm(torch.mean(feat_a, dim=1), p=2, dim=1))
loss_nor = torch.norm(torch.mean(feat_n, dim=1), p=2, dim=1)

loss_um = torch.mean((loss_abn + loss_nor) ** 2)

Why you used the torch.abs and squared (loss_abn + loss_nor)? This don't match the equation. Maybe some trick to converge?

Regards!

Hi Sorry for the late reply. Doing this shows better convergence. it did the same thing as the equation in the paper.

from rtfm.

@tianyu0207 Can you give a reference to why this implementation is the same as Equation (6), or give proof?
Thanks a lot.

from rtfm.

Hi, I have the same question. Have you got any idea on this?

from rtfm.