search

vandit15 / Class-balanced-loss-pytorch

Pytorch implementation of the paper "Class-Balanced Loss Based on Effective Number of Samples"
MIT License
784 stars 120 forks source link

What is samples_per_cls? #5

Closed Bonsen closed 5 years ago

Bonsen commented 5 years ago

I should compute samples_per_cls of whole dataset or each batch? If there is 0 in samples_per_cls of each batch, the loss will be nan.

DecentMakeover commented 4 years ago

How did you solve this?

rahulvigneswaran commented 4 years ago

@DecentMakeover @Bonsen @vandit15 If am not wrong, it is the samples_per_cls for the whole dataset. This is so that we can use the n_y as a power for beta to calculate the effective number.

davidhuangal commented 3 years ago

@rahulvigneswaran So do you mean that if your dataset looked like this: Class 0: 100 images Class 1: 200 images Class 2: 60 images

You would always set samples_per_cls to [100,200,60] ?