How long will it take to train with DICEHD on multiorgan dataset by nnunet? about seglossodyssey HOT 2 CLOSED

Hi @Tunaaaaa ,

It's normal because both HD loss and BD loss need to compute the distance transform map that costs lots of CPU computations.

The right way to use the two losses is training with Dice loss and fine-tune with DiceHD/DiceBD loss. I would recommend looking at the original papers to get more insights:)

from seglossodyssey.

Hi @Tunaaaaa ,

It's normal because both HD loss and BD loss need to compute the distance transform map that costs lots of CPU computations.

The right way to use the two losses is training with Dice loss and fine-tune with DiceHD/DiceBD loss. I would recommend looking at the original papers to get more insights:)

Hi,

Thank you~ At least I didn't get the code wrong, which is good news.  XD

Best
Junyu

from seglossodyssey.