Hi Guys,
Do you have any recommendations for loss function in the Imbalanced multiclass Semantic Segmentation problem? The weight difference in each class is pretty huge actually, like [0.1, 0.1, 0.4, 0.4]
currently, I’m using LovaszBCELoss, it’s good for accuracy, but the f1-score is pretty bad, 0.9 acc but 0.7 F1 (weighted F1).