search

HiLab-git / ACELoss

Implementations of "Learning Euler's Elastica Model for Medical Image Segmentation"
MIT License
71 stars 11 forks source link

Question for use aceloss #5

Closed Shrekbaby closed 3 years ago

Shrekbaby commented 3 years ago

Hi, I use the FastACELoss3DV2 as loss function for liver segmentation in nnUNet. Training about 140 epochs but the global dice for validation data set always below 0.04. Is there any wrong with me in using the loss function? aceloss

Luoxd1996 commented 3 years ago

Hi, This work was done one and half years ago, I have forgotten some training details. Your log is so strange, maybe you need to check the implementation is right firstly, and then you can combine the ACELoss with dice or CE, a tip is using torch.mean() rather than torch.sum(). Best; Xiangde.