Is fastai suitable for multiclass segmentation tasks with unbalanced classes?


As far as I know, fastai is the most suitable framework for image classification / segmentation tasks. I have multiclass image segmentation task, but my classes are highly unbalanced. Since oversampling is useless here, are there any techniques in fastai to deal with such tasks? For instance, is it possible to implement IoU metric with class weights (without using low-level PyTorch functionality)?

1 Like