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

Why you randomly sample one label from 4 lables in load_LIDC_data.py? #26

Closed Wenhui-Zhang-5 closed 2 years ago

Wenhui-Zhang-5 commented 2 years ago

Hi,

The code is helpful, thanks. However, I noticed that you are randomly sampling one ground truth label out of 4 in your file - load_LIDC_data.py. I guess this is not what the original paper means. Could you explain a bit about why you did this? Any comment is helpful. Thank you!

Best Stella

stefanknegt commented 2 years ago

This is correct, because it happens in the __getitem__ function and not in the initialization. Every time a sample is passed through the network we compare the output with a different ground truth, so over time the model sees all ground truths for the same sample.