This is an implementation of the U-Net model for Keras. When using the default parameters this will be the same as the original architecture, except that this code will use padded convolutions. If you are aware of other differences, please contact me.
U-Net: Convolutional Networks for Image Segmentation (https://arxiv.org/abs/1505.04597)
- Keeping the inc_rate at 1 often helped against overfitting and didn't hurt performance
- Dropout up to 0.2 yielded good results
- FCN feels like a good improvement over original U-Net
- I should really try separable and dilated convolutions