Keras-unet

unet.py

import keras
from keras import backend as K
from keras.layers.convolutional import Conv2D, Conv2DTranspose
from keras.layers import Input, Dense, Activation
from keras.layers import concatenate  # functional interface
from keras.models import Model
from keras.layers.advanced_activations import LeakyReLU

N_INPUT = 512


def get_unet():
    n_ch_exps = [4, 5, 6, 6, 7, 7]
    kernels = (5, 5)

    if K.image_data_format() == 'channels_first':
        ch_axis = 1
        input_shape = (1, N_INPUT, N_INPUT)
    elif K.image_data_format() == 'channels_last':
        ch_axis = 3
        input_shape = (N_INPUT, N_INPUT, 1)

    inp = Input(shape=input_shape)
    encodeds = []

    # encoder
    enc = inp
    for l_idx, n_ch in enumerate(n_ch_exps):
        enc = Conv2D(2 ** n_ch, kernels,
                     strides=(2, 2), padding='same',
                     kernel_initializer='he_normal')(enc)
        enc = LeakyReLU(name='encoded_{}'.format(l_idx),
                        alpha=0.2)(enc)
        encodeds.append(enc)

    # decoder
    dec = enc
    decoder_n_chs = n_ch_exps[::-1][1:]
    for l_idx, n_ch in enumerate(decoder_n_chs):
        l_idx_rev = len(n_ch_exps) - l_idx - 2  #
        dec = Conv2DTranspose(2 ** n_ch, kernels,
                              strides=(2, 2), padding='same',
                              kernel_initializer='he_normal',
                              activation='relu',
                              name='decoded_{}'.format(l_idx))(dec)
        dec = concatenate([dec, encodeds[l_idx_rev]],
                          axis=ch_axis)

    outp = Conv2DTranspose(1, kernels,
                           strides=(2, 2), padding='same',
                           kernel_initializer='glorot_normal',
                           activation='sigmoid',
                           name='decoded_{}'.format(l_idx + 1))(dec)

    unet = Model(inputs=inp, outputs=outp)
    return unet


if __name__ == "__main__":
    model = get_unet()
    model.summary()