class balanced loss?
2533245542 opened this issue · comments
Ever considered implementing class balanced loss? I found your repo useful and would like to see it include more.
Sorry for being so late. I was busy with other things when this issue is created. Frankly, I do not have such a plan to add it because the implementation you posted above is good and well documented.
I tried it in sementic segmentation task of cityscapes, but observed a bit drop of miou. The model I used is deeplabv3+, and I simply replace cross entropy with this loss without other changings. Maybe I should be more careful to tune the training parameters.