def conv2_block(input, k=1, dropout=0.0):
init = input
channel_axis = 1 if K.image_dim_ordering() == "th" else -1
x = BatchRenormalization(axis=channel_axis,
momentum=0.1,
epsilon=1e-5,
gamma_init='uniform')(input)
x = Activation('relu')(x)
x = Convolution2D(32 * k, (3, 3),
padding='same',
kernel_initializer='he_normal',
use_bias=False)(x)
if dropout > 0.0: x = Dropout(dropout)(x)
x = BatchRenormalization(axis=channel_axis,
momentum=0.1,
epsilon=1e-5,
gamma_init='uniform')(x)
x = Activation('relu')(x)
x = Convolution2D(32 * k, (3, 3),
padding='same',
kernel_initializer='he_normal',
use_bias=False)(x)
m = Add()([init, x])
return m
def ___conv4_block(input, k=1, dropout=0.0, mode=0):
init = input
channel_axis = 1 if K.image_dim_ordering() == "th" else -1
# Check if input number of filters is same as 64 * k, else create convolution2d for this input
if K.image_dim_ordering() == "th":
if init._keras_shape[1] != 64 * k:
init = Convolution2D(64 * k, 1, 1, activation='linear', border_mode='same')(init)
else:
if init._keras_shape[-1] != 64 * k:
init = Convolution2D(64 * k, 1, 1, activation='linear', border_mode='same')(init)
x = Convolution2D(64 * k, 3, 3, border_mode='same')(input)
x = BatchRenormalization(axis=channel_axis, mode=mode)(x)
x = Activation('relu')(x)
if dropout > 0.0:
x = Dropout(dropout)(x)
x = Convolution2D(64 * k, 3, 3, border_mode='same')(x)
x = BatchRenormalization(axis=channel_axis, mode=mode)(x)
x = Activation('relu')(x)
m = merge([init, x], mode='sum')
return m
def expand_conv(init, base, k, strides=(1, 1)):
x = Convolution2D(base * k, (3, 3),
padding='same',
strides=strides,
kernel_initializer='he_normal',
use_bias=False)(init)
channel_axis = 1 if K.image_data_format() == "channels_first" else -1
x = BatchRenormalization(axis=channel_axis,
momentum=0.1,
epsilon=1e-5,
gamma_init='uniform')(x)
x = Activation('relu')(x)
x = Convolution2D(base * k, (3, 3),
padding='same',
kernel_initializer='he_normal',
use_bias=False)(x)
skip = Convolution2D(base * k, (1, 1),
padding='same',
strides=strides,
kernel_initializer='he_normal',
use_bias=False)(init)
m = Add()([x, skip])
return m
def __conv1_block(input, mode=0):
x = Convolution2D(16, 3, 3, border_mode='same')(input)
channel_axis = 1 if K.image_dim_ordering() == "th" else -1
x = BatchRenormalization(axis=channel_axis, mode=mode)(x)
x = Activation('relu')(x)
return x
def inceptionv3(img_dim):
input_tensor = Input(shape=img_dim)
base_model = InceptionV3(include_top=False,
weights='imagenet',
input_shape=img_dim)
bn = BatchRenormalization()(input_tensor)
x = base_model(bn)
x = GlobalAveragePooling2D()(x)
x = Dropout(0.5)(x)
output = Dense(1, activation='sigmoid')(x)
model = Model(input_tensor, output)
return model
def initial_conv(input):
x = Convolution2D(16, (3, 3),
padding='same',
kernel_initializer='he_normal',
use_bias=False)(input)
channel_axis = 1 if K.image_data_format() == "channels_first" else -1
x = BatchRenormalization(axis=channel_axis,
momentum=0.1,
epsilon=1e-5,
gamma_init='uniform')(x)
x = Activation('relu')(x)
return x
def BRDNet(): #original format def BRDNet(), data is used to obtain the reshape of input data
inpt = Input(shape=(None,None,1)) #if the image is 3, it is color image. If the image is 1, it is gray color, 201807082123tcw
# 1st layer, Conv+relu
x = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1), padding='same')(inpt)
x = BatchRenormalization(axis=-1, epsilon=1e-3)(x)
x = Activation('relu')(x)
# 15 layers, Conv+BN+relu
for i in range(7):
x = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1), padding='same')(x)
x = BatchRenormalization(axis=-1, epsilon=1e-3)(x)
x = Activation('relu')(x)
# last layer, Conv
for i in range(8):
x = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1), padding='same')(x)
x = BatchRenormalization(axis=-1, epsilon=1e-3)(x)
x = Activation('relu')(x)
x = Conv2D(filters=1, kernel_size=(3,3), strides=(1,1), padding='same')(x) #gray is 1 color is 3
x = Subtract()([inpt, x]) # input - noise
y = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1), padding='same')(inpt)
y = BatchRenormalization(axis=-1, epsilon=1e-3)(y)
y = Activation('relu')(y)
# 15 layers, Conv+BN+relu
for i in range(7):
y = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1),dilation_rate=(2,2), padding='same')(y)
y = Activation('relu')(y)
y = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1), padding='same')(y)
y = BatchRenormalization(axis=-1, epsilon=1e-3)(y)
y = Activation('relu')(y)
for i in range(6):
y = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1),dilation_rate=(2,2), padding='same')(y)
y = Activation('relu')(y)
y = Conv2D(filters=64, kernel_size=(3,3), strides=(1,1), padding='same')(y)
y = BatchRenormalization(axis=-1, epsilon=1e-3)(y)
y = Activation('relu')(y)
y = Conv2D(filters=1, kernel_size=(3,3), strides=(1,1), padding='same')(y)#gray is 1 color is 3
y = Subtract()([inpt, y]) # input - noise
o = concatenate([x,y],axis=-1)
z = Conv2D(filters=1, kernel_size=(3,3), strides=(1,1), padding='same')(o)#gray is 1 color is 3
z= Subtract()([inpt, z])
model = Model(inputs=inpt, outputs=z)
return model
def get_model(model_nr=1,
kernel_size=(3, 3),
pool_size=(4, 4),
first_filters=32,
second_filters=64,
conv=False):
# build the model
model = Sequential()
if model_nr == 1: #No normalization
model.add(
Conv2D(first_filters,
kernel_size,
activation='relu',
padding='same',
input_shape=(IMAGE_SIZE, IMAGE_SIZE, 3)))
model.add(MaxPool2D(pool_size=pool_size))
model.add(
Conv2D(second_filters,
kernel_size,
activation='relu',
padding='same'))
model.add(MaxPool2D(pool_size=pool_size))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
elif model_nr == 2: #Batch renormalization
model.add(
Conv2D(first_filters,
kernel_size,
padding='same',
input_shape=(IMAGE_SIZE, IMAGE_SIZE, 3)))
model.add(BatchRenormalization())
model.add(Activation("relu"))
model.add(MaxPool2D(pool_size=pool_size))
model.add(Conv2D(second_filters, kernel_size, padding='same'))
model.add(BatchRenormalization())
model.add(Activation("relu"))
model.add(MaxPool2D(pool_size=pool_size))
model.add(Flatten())
model.add(Dense(64, init='uniform'))
model.add(BatchRenormalization())
model.add(Activation("relu"))
model.add(Dense(1, init='uniform'))
model.add(BatchRenormalization())
model.add(Activation("sigmoid"))
# if model_nr == 3: #Batch Instance normalization
# model opbouw van het enige voorbeeld is totaal anders
# vb: http://easy-tensorflow.com/tf-tutorials/convolutional-neural-nets-cnns?view=category&id=91
model.compile(SGD(lr=0.01, momentum=0.95),
loss='binary_crossentropy',
metrics=['accuracy'])
return model
def get_unet_renorm_1280(input_shape=(1280, 1280, 3),
num_classes=1):
inputs = Input(shape=input_shape)
# 2048, 1280
down0c = Conv2D(4, (3, 3), padding='same')(inputs)
down0c = BatchRenormalization()(down0c)
down0c = Activation('relu')(down0c)
down0c = Conv2D(4, (3, 3), padding='same')(down0c)
down0c = BatchRenormalization()(down0c)
down0c = Activation('relu')(down0c)
down0c_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0c)
# 1024, 640
down0b = Conv2D(8, (3, 3), padding='same')(down0c_pool)
down0b = BatchRenormalization()(down0b)
down0b = Activation('relu')(down0b)
down0b = Conv2D(8, (3, 3), padding='same')(down0b)
down0b = BatchRenormalization()(down0b)
down0b = Activation('relu')(down0b)
down0b_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0b)
# 512, 320
down0a = Conv2D(16, (3, 3), padding='same')(down0b_pool)
down0a = BatchRenormalization()(down0a)
down0a = Activation('relu')(down0a)
down0a = Conv2D(16, (3, 3), padding='same')(down0a)
down0a = BatchRenormalization()(down0a)
down0a = Activation('relu')(down0a)
down0a_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0a)
# 256, 160
down0 = Conv2D(32, (3, 3), padding='same')(down0a_pool)
down0 = BatchRenormalization()(down0)
down0 = Activation('relu')(down0)
down0 = Conv2D(32, (3, 3), padding='same')(down0)
down0 = BatchRenormalization()(down0)
down0 = Activation('relu')(down0)
down0_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0)
# 128, 80
down1 = Conv2D(64, (3, 3), padding='same')(down0_pool)
down1 = BatchRenormalization()(down1)
down1 = Activation('relu')(down1)
down1 = Conv2D(64, (3, 3), padding='same')(down1)
down1 = BatchRenormalization()(down1)
down1 = Activation('relu')(down1)
down1_pool = MaxPooling2D((2, 2), strides=(2, 2))(down1)
# 64, 40
down2 = Conv2D(128, (3, 3), padding='same')(down1_pool)
down2 = BatchRenormalization()(down2)
down2 = Activation('relu')(down2)
down2 = Conv2D(128, (3, 3), padding='same')(down2)
down2 = BatchRenormalization()(down2)
down2 = Activation('relu')(down2)
down2_pool = MaxPooling2D((2, 2), strides=(2, 2))(down2)
# 32, 20
down3 = Conv2D(256, (3, 3), padding='same')(down2_pool)
down3 = BatchRenormalization()(down3)
down3 = Activation('relu')(down3)
down3 = Conv2D(256, (3, 3), padding='same')(down3)
down3 = BatchRenormalization()(down3)
down3 = Activation('relu')(down3)
down3_pool = MaxPooling2D((2, 2), strides=(2, 2))(down3)
# 16, 10
down4 = Conv2D(512, (3, 3), padding='same')(down3_pool)
down4 = BatchRenormalization()(down4)
down4 = Activation('relu')(down4)
down4 = Conv2D(512, (3, 3), padding='same')(down4)
down4 = BatchRenormalization()(down4)
down4 = Activation('relu')(down4)
down4_pool = MaxPooling2D((2, 2), strides=(2, 2))(down4)
# 8, 5
center = Conv2D(1024, (3, 3), padding='same')(down4_pool)
center = BatchRenormalization()(center)
center = Activation('relu')(center)
center = Conv2D(1024, (3, 3), padding='same')(center)
center = BatchRenormalization()(center)
center = Activation('relu')(center)
# center
up4 = UpSampling2D((2, 2))(center)
up4 = concatenate([down4, up4], axis=3)
up4 = Conv2D(512, (3, 3), padding='same')(up4)
up4 = BatchRenormalization()(up4)
up4 = Activation('relu')(up4)
up4 = Conv2D(512, (3, 3), padding='same')(up4)
up4 = BatchRenormalization()(up4)
up4 = Activation('relu')(up4)
up4 = Conv2D(512, (3, 3), padding='same')(up4)
up4 = BatchRenormalization()(up4)
up4 = Activation('relu')(up4)
# 16
up3 = UpSampling2D((2, 2))(up4)
up3 = concatenate([down3, up3], axis=3)
up3 = Conv2D(256, (3, 3), padding='same')(up3)
up3 = BatchRenormalization()(up3)
up3 = Activation('relu')(up3)
up3 = Conv2D(256, (3, 3), padding='same')(up3)
up3 = BatchRenormalization()(up3)
up3 = Activation('relu')(up3)
up3 = Conv2D(256, (3, 3), padding='same')(up3)
up3 = BatchRenormalization()(up3)
up3 = Activation('relu')(up3)
# 32
up2 = UpSampling2D((2, 2))(up3)
up2 = concatenate([down2, up2], axis=3)
up2 = Conv2D(128, (3, 3), padding='same')(up2)
up2 = BatchRenormalization()(up2)
up2 = Activation('relu')(up2)
up2 = Conv2D(128, (3, 3), padding='same')(up2)
up2 = BatchRenormalization()(up2)
up2 = Activation('relu')(up2)
up2 = Conv2D(128, (3, 3), padding='same')(up2)
up2 = BatchRenormalization()(up2)
up2 = Activation('relu')(up2)
# 64
up1 = UpSampling2D((2, 2))(up2)
up1 = concatenate([down1, up1], axis=3)
up1 = Conv2D(64, (3, 3), padding='same')(up1)
up1 = BatchRenormalization()(up1)
up1 = Activation('relu')(up1)
up1 = Conv2D(64, (3, 3), padding='same')(up1)
up1 = BatchRenormalization()(up1)
up1 = Activation('relu')(up1)
up1 = Conv2D(64, (3, 3), padding='same')(up1)
up1 = BatchRenormalization()(up1)
up1 = Activation('relu')(up1)
# 128
up0 = UpSampling2D((2, 2))(up1)
up0 = concatenate([down0, up0], axis=3)
up0 = Conv2D(32, (3, 3), padding='same')(up0)
up0 = BatchRenormalization()(up0)
up0 = Activation('relu')(up0)
up0 = Conv2D(32, (3, 3), padding='same')(up0)
up0 = BatchRenormalization()(up0)
up0 = Activation('relu')(up0)
up0 = Conv2D(32, (3, 3), padding='same')(up0)
up0 = BatchRenormalization()(up0)
up0 = Activation('relu')(up0)
# 256
up0a = UpSampling2D((2, 2))(up0)
up0a = concatenate([down0a, up0a], axis=3)
up0a = Conv2D(16, (3, 3), padding='same')(up0a)
up0a = BatchRenormalization()(up0a)
up0a = Activation('relu')(up0a)
up0a = Conv2D(16, (3, 3), padding='same')(up0a)
up0a = BatchRenormalization()(up0a)
up0a = Activation('relu')(up0a)
up0a = Conv2D(16, (3, 3), padding='same')(up0a)
up0a = BatchRenormalization()(up0a)
up0a = Activation('relu')(up0a)
# 512
up0b = UpSampling2D((2, 2))(up0a)
up0b = concatenate([down0b, up0b], axis=3)
up0b = Conv2D(8, (3, 3), padding='same')(up0b)
up0b = BatchRenormalization()(up0b)
up0b = Activation('relu')(up0b)
up0b = Conv2D(8, (3, 3), padding='same')(up0b)
up0b = BatchRenormalization()(up0b)
up0b = Activation('relu')(up0b)
up0b = Conv2D(8, (3, 3), padding='same')(up0b)
up0b = BatchRenormalization()(up0b)
up0b = Activation('relu')(up0b)
# 1024
up0c = UpSampling2D((2, 2))(up0b)
up0c = concatenate([down0c, up0c], axis=3)
up0c = Conv2D(8, (3, 3), padding='same')(up0c)
up0c = BatchRenormalization()(up0c)
up0c = Activation('relu')(up0c)
up0c = Conv2D(8, (3, 3), padding='same')(up0c)
up0c = BatchRenormalization()(up0c)
up0c = Activation('relu')(up0c)
up0c = Conv2D(8, (3, 3), padding='same')(up0c)
up0c = BatchRenormalization()(up0c)
up0c = Activation('relu')(up0c)
# 2048
classify = Conv2D(num_classes, (1, 1), activation='sigmoid')(up0c)
model = Model(inputs=inputs, outputs=classify)
model.compile(optimizer=RMSprop(lr=0.0001), loss=bce_dice_loss, metrics=[dice_loss])
return model
def create_wide_residual_network(input_dim,
nb_classes=100,
N=2,
k=1,
dropout=0.0,
verbose=1):
"""
Creates a Wide Residual Network with specified parameters
:param input: Input Keras object
:param nb_classes: Number of output classes
:param N: Depth of the network. Compute N = (n - 4) / 6.
Example : For a depth of 16, n = 16, N = (16 - 4) / 6 = 2
Example2: For a depth of 28, n = 28, N = (28 - 4) / 6 = 4
Example3: For a depth of 40, n = 40, N = (40 - 4) / 6 = 6
:param k: Width of the network.
:param dropout: Adds dropout if value is greater than 0.0
:param verbose: Debug info to describe created WRN
:return:
"""
channel_axis = 1 if K.image_data_format() == "channels_first" else -1
ip = Input(shape=input_dim)
x = initial_conv(ip)
nb_conv = 4
x = expand_conv(x, 16, k)
for i in range(N - 1):
x = conv1_block(x, k, dropout)
nb_conv += 2
x = BatchRenormalization(axis=channel_axis,
momentum=0.1,
epsilon=1e-5,
gamma_init='uniform')(x)
x = Activation('relu')(x)
x = expand_conv(x, 32, k, strides=(2, 2))
for i in range(N - 1):
x = conv2_block(x, k, dropout)
nb_conv += 2
x = BatchRenormalization(axis=channel_axis,
momentum=0.1,
epsilon=1e-5,
gamma_init='uniform')(x)
x = Activation('relu')(x)
x = expand_conv(x, 64, k, strides=(2, 2))
for i in range(N - 1):
x = conv3_block(x, k, dropout)
nb_conv += 2
x = AveragePooling2D((8, 8))(x)
x = Flatten()(x)
x = Dense(nb_classes, activation='softmax')(x)
model = Model(ip, x)
if verbose: print("Wide Residual Network-%d-%d created." % (nb_conv, k))
return model