hanxunh / active-passive-losses Goto Github PK

Thanks for your brilliant work!
I've been working on semantic segmentation with noisy labels recently. Have you tried this idea on any segmentation datasets? Or what changes can be done for apl-losses to adapt to segmentation tasks?

Hi there,

Thank you for your fantastic work. It seems like none of the loss functions has the weight argument. Is it possible to apply the weight balanced loss for some unbalanced noisy datasets?

Kind regards,

Yuning Zhou

Hi,
thanks for sharing your implementation. I have two questions about it:

1. Does it also work on tabular data?
2. Is it possible to identify the noisy instances (return the noisy IDs or the clean set)?

Thanks!

Hi, thank you for your interesting work, but I had some confusion about the principle of active-passive loss when reading the paper. For passive loss such as MAE and RCE they can be transformed into A(1-p (y|x)) through the condition that the sum of probabilities is 1. By definition, it is also an active loss, so why can they combine with NCE to achieve amazing good results, But other active + active loss effects are not good. How do you understand it