This is a U-Net implementation as described in Ronneberger et al. using PyTorch. While the code was used to train on density images, it can be used for image segmentation by replacing the loss function.
The network is defined in unet.py and training is done in train.py.