def identity(x,
filters,
weights_decay=0.,
kernel_initializer='he_uniform',
use_bias=False,
bottleneck=2,
activation=lambda: Activation('relu')):
conv = Conv3D(filters // bottleneck,
kernel_size=(1, 1, 1),
padding='same',
use_bias=use_bias,
kernel_initializer=kernel_initializer,
kernel_regularizer=l2_penality(weights_decay))(x)
bn = BatchNormalization(scale=False, axis=-1)(conv)
conv = activation()(bn)
conv = Conv3D(filters // bottleneck,
kernel_size=(3, 3, 3),
padding='same',
use_bias=use_bias,
kernel_initializer=kernel_initializer,
kernel_regularizer=l2_penality(weights_decay))(conv)
bn = BatchNormalization(scale=False, axis=-1)(conv)
conv = activation()(bn)
conv = Conv3D(filters,
kernel_size=(1, 1, 1),
padding='same',
use_bias=use_bias,
kernel_initializer=kernel_initializer,
kernel_regularizer=l2_penality(weights_decay))(conv)
bn = BatchNormalization(scale=False, axis=-1)(conv)
residual = add([x, bn])
relu = activation()(residual)
return relu
def Reduction_A(x, filters, weights_decay=0., kernel_initializer='he_uniform', use_bias=False, bottleneck=2, activation=lambda: Activation('relu')):
#down1
down1 = Conv3D(filters//bottleneck, kernel_size=(1,1,1), strides=(2,2,2), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(x)
down1 = BatchNormalization(scale=False, axis=-1)(down1)
down1 = activation()(down1)
#down2
down2 = MaxPooling3D((3, 3, 3), strides=(2, 2, 2),padding='same')(x)
#down3
down3 = Conv3D(filters//bottleneck, kernel_size=(1,1,1), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(x)
down3 = BatchNormalization(scale=False, axis=-1)(down3)
down3 = activation()(down3)
down3 = Conv3D(filters//bottleneck, kernel_size=(3,3,3), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(down3)
down3 = BatchNormalization(scale=False, axis=-1)(down3)
down3 = activation()(down3)
down3 = Conv3D(filters, kernel_size=(3,3,3), strides=(2,2,2), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(down3)
down3 = BatchNormalization(scale=False, axis=-1)(down3)
down3 = activation()(down3)
down_mixed = concatenate([down1,down2,down3],axis=-1)
down_mixed = Conv3D(filters, kernel_size=(1,1,1), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(down_mixed)
down_mixed = BatchNormalization(scale=False, axis=-1)(down_mixed)
down_mixed = activation()(down_mixed)
return down_mixed
def get_model(dhw=[48,48,48],weights_decay = 0.,kernel_initializer='he_uniform',weights=None, deeper=False, activation=lambda: Activation('relu')):
shape = dhw+[1]
inputs = Input(shape = shape)
conv1 = Conv3D(32, kernel_size=(3, 3, 3), padding='same', use_bias=False, kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(inputs)
id1 = identity_resnet_A(conv1, filters=32, weights_decay=weights_decay, kernel_initializer=kernel_initializer,
use_bias=False, bottleneck=2,activation=activation)
down1 = Reduction_A(id1, filters=64, weights_decay=weights_decay, kernel_initializer=kernel_initializer,
use_bias=False, bottleneck=2,activation=activation)
id2 = identity_resnet_A(down1, filters=64, weights_decay=weights_decay, kernel_initializer=kernel_initializer,
use_bias=False, bottleneck=2,activation=activation)
down2 = Reduction_A(id2, filters=128, weights_decay=weights_decay, kernel_initializer=kernel_initializer,
use_bias=False, bottleneck=2,activation=activation)
id3 = identity_resnet_A(down2, filters=128, weights_decay=weights_decay, kernel_initializer=kernel_initializer,
use_bias=False, bottleneck=2,activation=activation)
down3 = Reduction_A(id3, filters=256, weights_decay=weights_decay, kernel_initializer=kernel_initializer,
use_bias=False, bottleneck=2,activation=activation)
pool = GlobalAvgPool3D()(down3)
outputs = Dense(1, kernel_regularizer=l2_penality(weights_decay), kernel_initializer=kernel_initializer, activation='sigmoid')(pool)
model = Model(inputs, outputs)
if weights is not None:
model.load_weights(weights, by_name=True)
return model
def identity_resnet_A(x, filters, weights_decay=0., kernel_initializer='he_uniform', use_bias=False, bottleneck=2, activation=lambda: Activation('relu')):
#channel1
conv_channel1 = Conv3D(filters//bottleneck, kernel_size=(1,1,1), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(x)
conv_channel1 = BatchNormalization(scale=False, axis=-1)(conv_channel1)
conv_channel1 = activation()(conv_channel1)
#channel2
conv_channel2 = Conv3D(filters//bottleneck, kernel_size=(1,1,1), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(x)
conv_channel2 = BatchNormalization(scale=False, axis=-1)(conv_channel2)
conv_channel2 = activation()(conv_channel2)
conv_channel2 = Conv3D(filters//bottleneck, kernel_size=(3,3,3), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(conv_channel2)
conv_channel2 = BatchNormalization(scale=False, axis=-1)(conv_channel2)
conv_channel2 = activation()(conv_channel2)
#channel3
conv_channel3 = Conv3D(filters//bottleneck, kernel_size=(1,1,1), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(x)
conv_channel3 = BatchNormalization(scale=False, axis=-1)(conv_channel3)
conv_channel3 = activation()(conv_channel3)
conv_channel3 = Conv3D(filters//bottleneck, kernel_size=(3,3,3), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(conv_channel3)
conv_channel3 = BatchNormalization(scale=False, axis=-1)(conv_channel3)
conv_channel3 = activation()(conv_channel3)
conv_channel3 = Conv3D(filters//bottleneck, kernel_size=(3,3,3), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(conv_channel3)
conv_channel3 = BatchNormalization(scale=False, axis=-1)(conv_channel3)
conv_channel3 = activation()(conv_channel3)
concat1 = concatenate([conv_channel1,conv_channel2,conv_channel3],axis=-1)
conv_channel123 = Conv3D(filters, kernel_size=(1,1,1), padding='same', use_bias=use_bias,
kernel_initializer=kernel_initializer,kernel_regularizer=l2_penality(weights_decay))(concat1)
conv_channel123 = BatchNormalization(scale=False, axis=-1)(conv_channel123)
residual = add([x, conv_channel123])
relu = activation()(residual)
return relu