[Question] What is filter g?

Question

[Question] What is filter g?

Opened this issue a year ago · 6 comments

Thanks for your great work!
In your paper pages 7 and 8, you mention filters $g$ and $\bar g$.
What exactly is this filter, is it a network module? How did you get it?

Answer 1 · 2023-10-10T16:25:00.000Z

it's just a simple selection operation to calculate the related label possibilities. For example, for fc output: [0.1,0.2,0.2,0.5], if we want to preserve the first two labels, g([0.1,0.2,0.2,0.5]) = [0.1,0.2]

Answer 2 · 2023-10-11T03:31:46.000Z

Thanks for your quick reply.
I have another question.
In https://github.com/JngwenYe/LIRF/blob/main/train_deposit.py#L150-L158 , the code does not use criterionAT(output_batch1, output_teacher_batch1.detach()), is the current version of code use our LIRF method?

which loss do loss1, loss2 and loss3 correspond to in the paper ($L_{kp}$ etc.)?

Answer 3 · 2023-10-11T03:36:59.000Z

And for the return value in https://github.com/JngwenYe/LIRF/blob/main/train_deposit.py#L174, are two return values the same?

Answer 4 · 2023-10-13T05:53:11.000Z

@JngwenYe Can you share the complete code of your work? Thanks

Answer 5 · 2023-12-11T08:38:30.000Z

@JngwenYe Can you share the complete code of your work? Thanks
Did you successfully reproduce it

Answer 6 · 2023-12-11T10:50:18.000Z

@JngwenYe Can you share the complete code of your work? Thanks
Did you successfully reproduce it

no