search

jiawei-ren / BalancedMetaSoftmax-Classification

[NeurIPS 2020] Balanced Meta-Softmax for Long-Tailed Visual Recognition
https://github.com/jiawei-ren/BalancedMetaSoftmax
Other
135 stars 26 forks source link

I have one question about sample_per_class in balanced_softmax_loss #15

Closed Bhongcheng closed 1 year ago

Bhongcheng commented 1 year ago

Dear friend, I wonder what would happen if there are "0" in sample_per_class(missing class), as 0.log() is -inf. In that case, what is the meaning of it? Could you please inform me?

jiawei-ren commented 1 year ago

If there are unseen classes with 0 samples in training, it will be a zero-shot problem where our method is no longer applicable.