Does this loss work in multiclass problem?

Digital-Dermatology / t-loss

Official code for Robust T-Loss for Medical Image Segmentation (MICCAI 2023)

https://robust-tloss.github.io/

Apache License 2.0

41 stars 6 forks source link

Does this loss work in multiclass problem? #5

Open Marcelo5444 opened 3 months ago

alvarogonjim commented 2 months ago

Hello @Marcelo5444, We have not tested this loss in a multi-class setting. Since we have reparametrized the covariance matrix, I suspect that the loss might be unstable and not provide good results. However, feel free to give it a try and share your results in this issue. It will be valuable for the community.