Keras implementation of a 2D/3D U-Net with the following implementations provided:

• Additive attention -- Attention U-Net: Learning Where to Look for the Pancreas
• Inception convolutions w/ dilated convolutions -- Going Deeper with Convolutions and Multi-Scale Context Aggregation by Dilated Convolutions
• Recurrent convolutions -- R2U-Net
• Focal Tversky Loss
• Dice Coefficient Loss

This repository depends on the following libraries:

• Tensorflow
• Keras
• Python 3
• Numpy
• Matplotlib

The pre-implemented layers are available in layers3D.py. Use the layers to build your preferred network configuration in network.py

from layers3D import *
from tensorflow.keras.models import Model

def network(input_img, n_filters=16, dropout=0.5, batchnorm=True):
    outputs = inception_block(input_img, n_filters=n_filters, batchnorm=batchnorm, strides=1, recurrent=2)
    model = Model(inputs=[input_img], outputs=[outputs])
    return model

Refer to network.py for a full example

Rewrite the __data_generation() method in datagenerator.py to supply batches of data during training

def __data_generation(self, list_IDs_temp):

        X = np.empty((self.batch_size, *self.dim, self.n_channels))
        y = np.empty((self.batch_size, *self.dim, self.n_channels))

        for i, ID in enumerate(list_IDs_temp):
            # Write logic for selecting/manipulating X and y here
            X[i,] = np.load('path/to/x/ID')
            y[i,] = np.load('path/to/y/ID')

        return X, y

The DataGenerator class in train.py takes in list arguments containing the ID (filenames) of X and y

Set the appropriate values for the hyper-parameters listed in hyperparameters.py

Run train.py once all the configuration is done to train your network

Run evaluate.py or predict.py with the appropriate list_IDs provided to the DataGenerator