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

Unet features #12

Closed schenock closed 3 years ago

schenock commented 3 years ago

The U net seems that it is not being trained (which is in accordance to the paper). However, you do not load any weights (pretrained model) into your U Net model as far as I was able to see. This means that unet weights are as initialized.

Is this a correct observation? Was the purpose of the code just showcasing the implementation of Probabilistic UNet, because I assume it will be difficult to train without a trained segmentation network.

stefanknegt commented 3 years ago

We do train the Unet. The Unet is part of the Probabilistic Unet, so they are both being trained.

schenock commented 3 years ago

ah got it. thanks