Leaky V-Net in Pytorch

This is a fork of @Dootmaan's VNet.PyTorch repo, who attempted a faithful recreation of the original V-Net: Fully Convolutional Neural Network for Volumetric Medical Image paper, with as little adaptations as possible.

This repo's model has some minor adaptations to fit it's designated application:

ReLU layers is now the leaky version to allow for more consistent convergence on small training datasets
Final Sigmoid layer has been removed from the network in favour of manual thresholding (for flexibility during testing)