search

stefanknegt / Probabilistic-Unet-Pytorch

A Probabilistic U-Net for segmentation of ambiguous images implemented in PyTorch
Apache License 2.0
270 stars 67 forks source link

sample #17

Closed ZhouCX117 closed 3 years ago

ZhouCX117 commented 3 years ago

Hi,could you tell me the sampling process? The sample function in the file(Probablistic_unet.py) did not been used. Thank you~

stefanknegt commented 3 years ago

Could you elaborate a bit? What do you want to know exactly?

ZhouCX117 commented 3 years ago

During the training process, I always get the Nan loss after some steps. Could you give me a suggestion?

stefanknegt commented 3 years ago

You should first find out which term is causing the loss to become NaN. I guess it’s the KL term. Then find out why it becomes NaN. I didn’t have this issue, so I guess it has something to do with your data.

ZhouCX117 commented 3 years ago

Thanks for your reply~ Could you tell me that what kind of uncertainty Probablistic UNet models? Data uncertainty or model uncertainty? Thanks a lot!

stefanknegt commented 3 years ago

For questions regarding the architecture you should really ask the original authors.