Open Easonrust opened 5 years ago
when the target area of the segmentation is particularly small on the whole picture, resulting in a serious imbalance of samples, so we change the loss function to ‘dice coef loss’
when the target area of the segmentation is particularly small on the whole picture, resulting in a serious imbalance of samples, so we change the loss function to ‘dice coef loss’