how to use

[xiaoshuyun]

Thank you for your excellent work.
I want to know, how can I use your seesaw loss to replace cross entropy in the traditional detection algorithm?
Looking forward to your reply.

[bamps53]

I believe you can use it by just replacing BCEWithLogitsLoss in your code with this variant.

SeesawLoss/seesaw_loss.py

Line 48 in 476d60f

class DistibutionAgnosticSeesawLossWithLogits(nn.Module):

But please note that it's now obsolete as original authors updated from 1st revision at ECCV 2020 workshop.
(It's mentioned in this thread)

[xiaoshuyun]

Can BCEWithLogitsLoss be used for multi-class classification or target detection ?

[kuotunyu]

Can BCEWithLogitsLoss be used for multi-class classification or target detection ?

Did you solve it?
I have absolutely no idea how to use it

In multi-class classification, I changed nn.CrossEntropyLoss() to DistibutionAgnosticSeesawLossWithLogits or SeesawLossWithLogits, but it keeps getting IndexError: too many indices for tensor of dimension 1 ....