About facal loss?

[rongyixiao]

Hello, does facal loss address data imbalance? if data is balance, is facal loss useless? In this case, loss in the original paper should be adopted?

[rongyixiao]

1 、Facal loss is supposed to increase learning of difficult samples, right? 2、Why is my facal loss almost the same? Thank you

[StevenBanama]

Hello, does facal loss address data imbalance? if data is balance, is facal loss useless? In this case, loss in the original paper should be adopted? @rongyixiao Q1:due to the unbalance distribution of wiki and imdb. Q2: you can use CE, when the dataset distribute balancely. Q3: origin papper may randomly select sample so that decrease the risk of overfitting and unbalance

[StevenBanama]

focal loss