The score is 0.996576 (dice coefficient) in private leader board. The best is 0.997332.
The neural network is trained using u-net-v5.ipynb.
The data_utils contains all codes to prepare data and perform data augmentation.
The keras_custom implement custom metric and loss. (dice coefficient and boundary penalized loss)
Overall network architecture is as follow:
