shruti-jadon / Semantic-Segmentation-Loss-Functions

This Repository is implementation of majority of Semantic Segmentation Loss Functions
MIT License
553 stars 84 forks source link

Question about constant values #6

Open vincentallen852 opened 2 years ago

vincentallen852 commented 2 years ago

Hello. Thank you for these loss functions they are very helpful.

I am curious about how to use the constant terms (alpha, beta, gamma, and epsilon) to factor in class imbalances. I am working on a data set with two classes [0 and 1] and around 99% of points are 0. Should I change the alpha and beta values to very small/ large values to reflect this?

vincentallen852 commented 2 years ago

I realize I should specify that I am using weighted_cross_entropyloss, focal_tversky, and log_cosh_dice_loss. Thank you.