def build_encoder_decoder():
# Encoder
input_tensor = Input(shape=(320, 320, 4))
x = ZeroPadding2D((1, 1))(input_tensor)
x = Conv2D(64, (3, 3), activation='relu', name='conv1_1')(x)
x = BatchNormalization()(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(64, (3, 3), activation='relu', name='conv1_2')(x)
x = BatchNormalization()(x)
orig_1 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='conv2_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='conv2_2')(x)
orig_2 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_3')(x)
orig_3 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
inputs_size = x.get_shape()[1:3]
conv_4_1x1 = Conv2D(512, (1, 1),
activation='relu',
padding='same',
name='conv4_1x1')(x)
conv_4_3x3_1 = Conv2D(512, (3, 3),
activation='relu',
padding='same',
dilation_rate=ATROUS_RATES[0],
name='conv4_3x3_1')(x)
conv_4_3x3_2 = Conv2D(512, (3, 3),
activation='relu',
padding='same',
dilation_rate=ATROUS_RATES[1],
name='conv4_3x3_2')(x)
conv_4_3x3_3 = Conv2D(512, (3, 3),
activation='relu',
padding='same',
dilation_rate=ATROUS_RATES[2],
name='conv4_3x3_3')(x)
# Image average pooling
image_level_features = Lambda(
lambda x: tf.reduce_mean(x, [1, 2], keepdims=True),
name='global_average_pooling')(x)
image_level_features = Conv2D(
512, (1, 1),
activation='relu',
padding='same',
name='image_level_features_conv_1x1')(image_level_features)
image_level_features = Lambda(lambda x: tf.image.resize(x, inputs_size),
name='upsample_1')(image_level_features)
# Concat
x = Concatenate(axis=3)([
conv_4_1x1, conv_4_3x3_1, conv_4_3x3_2, conv_4_3x3_3,
image_level_features
])
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
name='conv_1x1_1_concat')(x)
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
name='conv_1x1_2_concat')(x)
orig_4 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_3')(x)
orig_5 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
# Decoder
#
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_5)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_5)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(512, (3, 3),
activation='relu',
padding='same',
name='deconv5_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (3, 3),
activation='relu',
padding='same',
name='deconv5_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (3, 3),
activation='relu',
padding='same',
name='deconv5_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_4)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_4)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='deconv4_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='deconv4_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='deconv4_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_3)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_3)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='deconv3_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='deconv3_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='deconv3_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_2)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_2)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv2_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv2_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_1)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_1)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv1_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv1_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(1, (3, 3),
activation='sigmoid',
padding='same',
name='pred',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=input_tensor, outputs=x)
return model
def build_encoder_decoder():
num_labels = 8
kernel = 3
# Encoder
input_tensor = Input(shape=(320, 320, 3))
x = ZeroPadding2D((1, 1))(input_tensor)
x = Conv2D(64, (kernel, kernel), activation='relu', name='conv1_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(64, (kernel, kernel), activation='relu', name='conv1_2')(x)
orig_1 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (kernel, kernel), activation='relu', name='conv2_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (kernel, kernel), activation='relu', name='conv2_2')(x)
orig_2 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (kernel, kernel), activation='relu', name='conv3_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (kernel, kernel), activation='relu', name='conv3_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (kernel, kernel), activation='relu', name='conv3_3')(x)
orig_3 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (kernel, kernel), activation='relu', name='conv4_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (kernel, kernel), activation='relu', name='conv4_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (kernel, kernel), activation='relu', name='conv4_3')(x)
orig_4 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (kernel, kernel), activation='relu', name='conv5_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (kernel, kernel), activation='relu', name='conv5_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (kernel, kernel), activation='relu', name='conv5_3')(x)
orig_5 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
# Decoder
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_5)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_5)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(512, (kernel, kernel),
activation='relu',
padding='same',
name='deconv5_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel),
activation='relu',
padding='same',
name='deconv5_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel),
activation='relu',
padding='same',
name='deconv5_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_4)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_4)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (kernel, kernel),
activation='relu',
padding='same',
name='deconv4_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel),
activation='relu',
padding='same',
name='deconv4_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel),
activation='relu',
padding='same',
name='deconv4_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_3)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_3)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(128, (kernel, kernel),
activation='relu',
padding='same',
name='deconv3_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (kernel, kernel),
activation='relu',
padding='same',
name='deconv3_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (kernel, kernel),
activation='relu',
padding='same',
name='deconv3_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_2)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_2)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (kernel, kernel),
activation='relu',
padding='same',
name='deconv2_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (kernel, kernel),
activation='relu',
padding='same',
name='deconv2_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_1)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_1)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (kernel, kernel),
activation='relu',
padding='same',
name='deconv1_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (kernel, kernel),
activation='relu',
padding='same',
name='deconv1_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(num_labels, (1, 1),
activation='softmax',
padding='valid',
name='pred',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=input_tensor, outputs=x)
return model
def build_encoder_decoder():
# Encoder
input_tensor = Input(shape=(320, 320, 4))
x = ZeroPadding2D((1, 1))(input_tensor)
x = Conv2D(64, (3, 3), activation='relu', name='conv1_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(64, (3, 3), activation='relu', name='conv1_2')(x)
orig_1 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='conv2_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='conv2_2')(x)
orig_2 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_3')(x)
orig_3 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv4_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv4_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv4_3')(x)
orig_4 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_3')(x)
orig_5 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
# Decoder
# x = Conv2D(4096, (7, 7), activation='relu', padding='valid', name='conv6')(x)
# x = BatchNormalization()(x)
# x = UpSampling2D(size=(7, 7))(x)
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
name='deconv6',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_5)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_5)
# print('origReshaped.shape: ' + str(K.int_shape(origReshaped)))
xReshaped = Reshape(shape)(x)
# print('xReshaped.shape: ' + str(K.int_shape(xReshaped)))
together = Concatenate(axis=1)([origReshaped, xReshaped])
# print('together.shape: ' + str(K.int_shape(together)))
x = Unpooling()(together)
x = Conv2D(512, (5, 5),
activation='relu',
padding='same',
name='deconv5',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_4)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_4)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (5, 5),
activation='relu',
padding='same',
name='deconv4',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_3)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_3)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(128, (5, 5),
activation='relu',
padding='same',
name='deconv3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_2)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_2)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (5, 5),
activation='relu',
padding='same',
name='deconv2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_1)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_1)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (5, 5),
activation='relu',
padding='same',
name='deconv1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(1, (5, 5),
activation='sigmoid',
padding='same',
name='pred',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=input_tensor, outputs=x)
return model
def build_encoder_decoder():
kernel = 3
# Encoder
#
input_tensor = Input(shape=(320, 320, 4))
input_tensor_shape = input_tensor.get_shape()[1:3]
# Entry flow
x = Conv2D(32, (3, 3), padding="same")(input_tensor)
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = Conv2D(64, (3, 3), padding="same")(x)
orig_1 = BatchNormalization()(x)
x = Activation("relu")(orig_1)
x = x = Conv2D(64, (3, 3), strides=(2, 2), padding="same")(x)
x = BatchNormalization()(x)
x = Activation("relu")(x)
x, orig_2 = res_downsample_xception_block(x, 128)
x, orig_3 = res_downsample_xception_block(x, 256, top_relu=True)
x, orig_4 = res_downsample_xception_block(x, 728, top_relu=True)
# Middle flow
for i in range(8):
x = res_xception_block(x, 728)
# Exit flow
res = Conv2D(1024, (1, 1), padding="same")(x)
res = BatchNormalization()(res)
x = Activation("relu")(x)
x = DepthwiseConv2D((3, 3), padding="same")(x)
x = BatchNormalization()(x)
x = Conv2D(728, (1, 1), padding="same")(x)
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = DepthwiseConv2D((3, 3), padding="same")(x)
x = BatchNormalization()(x)
x = Conv2D(1024, (1, 1), padding="same")(x)
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = DepthwiseConv2D((3, 3), padding="same")(x)
x = BatchNormalization()(x)
x = Conv2D(1024, (1, 1), padding="same")(x)
x = BatchNormalization()(x)
x = Add()([x, res])
x = DepthwiseConv2D((3, 3), padding="same")(x)
x = BatchNormalization()(x)
x = Conv2D(1536, (1, 1), padding="same")(x)
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = DepthwiseConv2D((3, 3), padding="same")(x)
x = BatchNormalization()(x)
x = Conv2D(1536, (1, 1), padding="same")(x)
x = BatchNormalization()(x)
x = Activation("relu")(x)
x = DepthwiseConv2D((3, 3), padding="same")(x)
x = BatchNormalization()(x)
x = Conv2D(2048, (1, 1), padding="same")(x)
x = BatchNormalization()(x)
x = Activation("relu")(x)
inputs_size = x.get_shape()[1:3]
# Atrous convolution
conv_4_1x1 = SeparableConv2D(256, (1, 1),
activation='relu',
padding='same',
name='conv4_1x1')(x)
conv_4_3x3_1 = SeparableConv2D(256, (kernel, kernel),
activation='relu',
padding='same',
dilation_rate=ATROUS_RATES[0],
name='conv4_3x3_1')(x)
conv_4_3x3_2 = SeparableConv2D(256, (kernel, kernel),
activation='relu',
padding='same',
dilation_rate=ATROUS_RATES[1],
name='conv4_3x3_2')(x)
conv_4_3x3_3 = SeparableConv2D(256, (kernel, kernel),
activation='relu',
padding='same',
dilation_rate=ATROUS_RATES[2],
name='conv4_3x3_3')(x)
# Image average pooling
image_level_features = Lambda(
lambda x: tf.reduce_mean(x, [1, 2], keepdims=True),
name='global_average_pooling')(x)
image_level_features = Conv2D(
256, (1, 1),
activation='relu',
padding='same',
name='image_level_features_conv_1x1')(image_level_features)
image_level_features = Lambda(
lambda x: tf.image.resize_bilinear(x, inputs_size),
name='upsample_1')(image_level_features)
# Concat
x = Concatenate(axis=3)([
conv_4_1x1, conv_4_3x3_1, conv_4_3x3_2, conv_4_3x3_3,
image_level_features
])
x = Conv2D(256, (1, 1),
activation='relu',
padding='same',
name='conv_1x1_1_concat')(x)
x = Conv2D(728, (1, 1),
activation='relu',
padding='same',
name='conv_1x1_2_concat')(x)
# Decoderg
#
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_4)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_4)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='deconv4_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='deconv4_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (3, 3),
activation='relu',
padding='same',
name='deconv4_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_3)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_3)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='deconv3_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='deconv3_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (3, 3),
activation='relu',
padding='same',
name='deconv3_3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_2)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_2)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv2_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv2_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_1)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_1)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv1_1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3),
activation='relu',
padding='same',
name='deconv1_2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(1, (3, 3),
activation='sigmoid',
padding='same',
name='pred',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=input_tensor, outputs=x)
return model
def create_model():
# Encoder
input_tensor = Input(shape=(320, 320, 4))
x = ZeroPadding2D((1, 1))(input_tensor)
x = Conv2D(64, (3, 3), activation='relu', name='conv1_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(64, (3, 3), activation='relu', name='conv1_2')(x)
orig_1 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='conv2_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='conv2_2')(x)
orig_2 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='conv3_3')(x)
orig_3 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv4_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv4_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv4_3')(x)
orig_4 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='conv5_3')(x)
orig_5 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
# Decoder
# x = Conv2D(4096, (7, 7), activation='relu', padding='valid', name='conv6')(x)
# x = BatchNormalization()(x)
# x = UpSampling2D(size=(7, 7))(x)
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
name='deconv6',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
x = Unpooling(orig_5, (20, 20, 512))(x)
x = Conv2D(512, (5, 5),
activation='relu',
padding='same',
name='deconv5',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
x = Unpooling(orig_4, (40, 40, 512))(x)
x = Conv2D(256, (5, 5),
activation='relu',
padding='same',
name='deconv4',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
x = Unpooling(orig_3, (80, 80, 256))(x)
x = Conv2D(128, (5, 5),
activation='relu',
padding='same',
name='deconv3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
x = Unpooling(orig_2, (160, 160, 128))(x)
x = Conv2D(64, (5, 5),
activation='relu',
padding='same',
name='deconv2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
x = Unpooling(orig_1, (320, 320, 64))(x)
x = Conv2D(64, (5, 5),
activation='relu',
padding='same',
name='deconv1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(1, (5, 5),
activation='sigmoid',
padding='same',
name='pred',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=input_tensor, outputs=x)
return model
def build_model():
# Encoder
img_input = Input(shape=(img_rows, img_cols, channel))
x = ZeroPadding2D((1, 1))(img_input)
x = Conv2D(64, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv1_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(64, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv1_2')(x)
orig_1 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv2_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv2_2')(x)
orig_2 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv3_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv3_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv3_3')(x)
orig_3 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv4_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv4_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv4_3')(x)
orig_4 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv5_1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv5_2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', kernel_initializer='he_normal', name='conv5_3')(x)
orig_5 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
model = Model(img_input, x)
# Loads ImageNet pre-trained data
ensure_vgg16_weights()
weights_path = 'models/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5'
model.load_weights(weights_path)
# Decoder
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_5)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_5)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_1',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_2',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_3',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_4)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_4)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv4_1',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv4_2',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv4_3',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_3)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_3)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv3_1',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv3_2',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (kernel, kernel), activation='relu', padding='same', name='deconv3_3',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_2)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_2)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(128, (kernel, kernel), activation='relu', padding='same', name='deconv2_1',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (kernel, kernel), activation='relu', padding='same', name='deconv2_2',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_1)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_1)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(64, (kernel, kernel), activation='relu', padding='same', name='deconv1_1',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (kernel, kernel), activation='relu', padding='same', name='deconv1_2',
kernel_initializer='he_normal')(x)
x = BatchNormalization()(x)
outputs = Conv2D(num_classes, (1, 1), activation='softmax', padding='valid', name='pred',
kernel_initializer='he_normal')(x)
model = Model(inputs=img_input, outputs=outputs, name="SegNet")
return model
def build_encoder_decoder(hh, ww):
# Encoder
base_model = VGG16(include_top=False,
weights=None,
input_tensor=None,
input_shape=(hh, ww, 4),
pooling=None)
migrate_model(base_model)
x = base_model.output
orig_5 = base_model.get_layer('block5_pool').output
orig_4 = base_model.get_layer('block4_pool').output
orig_3 = base_model.get_layer('block3_pool').output
orig_2 = base_model.get_layer('block2_pool').output
orig_1 = base_model.get_layer('block1_pool').output
# Decoder
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
name='deconv6',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 512))(x)
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
# low_feature = Conv2D(512, (3, 3), activation='relu', padding='same', kernel_initializer='he_normal',bias_initializer='zeros')(orig_5)
attention_feature = multiply([orig_5, x])
#concate
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
# x = UpSampling2D(size=(2, 2))(x)
#add scale transfer
x = Conv2D(2048, (1, 1), padding='same', use_bias=False)(x)
x = pixel_shuffler((2, 2))(x)
#block_2
x = Conv2D(512, (5, 5),
activation='relu',
padding='same',
name='deconv5',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 512))(x)
x = Conv2D(512, (1, 1),
activation='relu',
padding='same',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
# low_feature = Conv2D(512, (3, 3), activation='relu', padding='same', kernel_initializer='he_normal',
# bias_initializer='zeros')(orig_4)
attention_feature = multiply([orig_4, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
# x = UpSampling2D(size=(2, 2))(x)
x = pixel_shuffler((2, 2))(x)
# block_3
x = Conv2D(256, (5, 5),
activation='relu',
padding='same',
name='deconv4',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 256))(x)
x = Conv2D(256, (1, 1),
activation='relu',
padding='same',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
# low_feature = Conv2D(256, (3, 3), activation='relu', padding='same', kernel_initializer='he_normal',
# bias_initializer='zeros')(orig_3)
attention_feature = multiply([orig_3, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
# x = UpSampling2D(size=(2, 2))(x)
x = pixel_shuffler((2, 2))(x)
# block_4
x = Conv2D(128, (5, 5),
activation='relu',
padding='same',
name='deconv3',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 128))(x)
x = Conv2D(128, (1, 1),
activation='relu',
padding='same',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
# low_feature = Conv2D(128, (3, 3), activation='relu', padding='same', kernel_initializer='he_normal',
# bias_initializer='zeros')(orig_2)
attention_feature = multiply([orig_2, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
# x = UpSampling2D(size=(2, 2))(x)
x = pixel_shuffler((2, 2))(x)
# block_5
x = Conv2D(64, (5, 5),
activation='relu',
padding='same',
name='deconv2',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 64))(x)
x = Conv2D(64, (1, 1),
activation='relu',
padding='same',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
# low_feature = Conv2D(64, (3, 3), activation='relu', padding='same', kernel_initializer='he_normal',
# bias_initializer='zeros')(orig_1)
attention_feature = multiply([orig_1, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
# x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(256, (1, 1), padding='same', use_bias=False)(x)
x = pixel_shuffler((2, 2))(x)
# block_6
x = Conv2D(64, (5, 5),
activation='relu',
padding='same',
name='deconv1',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
x = Conv2D(1, (5, 5),
activation='sigmoid',
padding='same',
name='pred',
kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=base_model.input, outputs=x)
return model
def build_model():
# Encoder
img_input = Input(shape=(img_rows, img_cols, channel))
x = ZeroPadding2D((1, 1))(img_input)
x = Conv2D(64, (3, 3), activation='relu', name='block1_conv1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(64, (3, 3), activation='relu', name='block1_conv2')(x)
orig_1 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='block2_conv1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(128, (3, 3), activation='relu', name='block2_conv2')(x)
orig_2 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='block3_conv1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='block3_conv2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='block3_conv3')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(256, (3, 3), activation='relu', name='block3_conv4')(x)
orig_3 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block4_conv1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block4_conv2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block4_conv3')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block4_conv4')(x)
orig_4 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block5_conv1')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block5_conv2')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block5_conv3')(x)
x = ZeroPadding2D((1, 1))(x)
x = Conv2D(512, (3, 3), activation='relu', name='block5_conv4')(x)
orig_5 = x
x = MaxPooling2D((2, 2), strides=(2, 2))(x)
# Decoder
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_5)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_5)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_1',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_2',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_3',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv5_4',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_4)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_4)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv4_1',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(512, (kernel, kernel), activation='relu', padding='same', name='deconv4_2',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv4_3',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv4_4',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_3)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_3)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv3_1',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(256, (kernel, kernel), activation='relu', padding='same', name='deconv3_2',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (kernel, kernel), activation='relu', padding='same', name='deconv3_3',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (kernel, kernel), activation='relu', padding='same', name='deconv3_4',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = UpSampling2D(size=(2, 2))(x)
the_shape = K.int_shape(orig_2)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(orig_2)
xReshaped = Reshape(shape)(x)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(128, (kernel, kernel), activation='relu', padding='same', name='deconv2_1',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (kernel, kernel), activation='relu', padding='same', name='deconv2_2',
kernel_initializer='he_uniform')(x)
x = BatchNormalization()(x)
x = Conv2D(3, (1, 1), padding='same', name='pred', activation='sigmoid', kernel_initializer='he_uniform')(x)
outputs = x
model = Model(inputs=img_input, outputs=outputs)
ensure_vgg_weights()
model.load_weights('models/vgg19_notop.h5', by_name=True)
return model
def build_encoder_decoder(hh,ww):
# encoder
resNet = ResNet50(include_top=False,weights= None, input_tensor=None, input_shape=(hh,ww,4), pooling=None)
# migrate_model(resNet)
x = resNet.output
orig_5 = resNet.get_layer('activation_49').output
orig_5 = Conv2D(512, (1, 1), padding='same', kernel_initializer='he_normal', bias_initializer='zeros')(orig_5)
orig_4 = resNet.get_layer('activation_40').output
orig_4 = Conv2D(512, (1, 1), padding='same', kernel_initializer='he_normal', bias_initializer='zeros')(orig_4)
orig_3 = resNet.get_layer('activation_22').output
orig_3 = Conv2D(256, (1, 1), padding='same', kernel_initializer='he_normal', bias_initializer='zeros')(orig_3)
orig_2 = resNet.get_layer('activation_10').output
orig_2 = Conv2D(128, (1, 1), padding='same', kernel_initializer='he_normal', bias_initializer='zeros')(orig_2)
orig_1 = resNet.get_layer('activation_1').output
orig_1 = Conv2D(64, (1, 1), padding='same', kernel_initializer='he_normal', bias_initializer='zeros')(orig_1)
#decoder
#stage_one
x = Conv2D(512, (1, 1), activation='relu', padding='same', name='deconv6', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1,1,512))(x)
x = Conv2D(512, (1, 1), activation='relu', padding='same', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
attention_feature = multiply([orig_5, x])
#concate
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
#add scale transfer
x = Conv2D(2048, (1,1),padding='same',use_bias=False)(x)
x = pixel_shuffler((2,2))(x)
#stage_two
x = Conv2D(512, (5, 5), activation='relu', padding='same', name='deconv5', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 512))(x)
x = Conv2D(512, (1,1), activation='relu', padding='same', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
attention_feature = multiply([orig_4, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = pixel_shuffler((2, 2))(x)
# stage_three
x = Conv2D(256, (5, 5), activation='relu', padding='same', name='deconv4', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 256))(x)
x = Conv2D(256, (1,1), activation='relu', padding='same', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
attention_feature = multiply([orig_3, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = pixel_shuffler((2, 2))(x)
# stage_four
x = Conv2D(128, (5, 5), activation='relu', padding='same', name='deconv3', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 128))(x)
x = Conv2D(128, (1,1), activation='relu', padding='same', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
attention_feature = multiply([orig_2, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = pixel_shuffler((2, 2))(x)
# stage_five
x = Conv2D(64, (5, 5), activation='relu', padding='same', name='deconv2', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
high_feature = x
x = GlobalAveragePooling2D()(x)
x = Reshape((1, 1, 64))(x)
x = Conv2D(64, (1,1), activation='relu', padding='same', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
attention_feature = multiply([orig_1, x])
the_shape = K.int_shape(attention_feature)
shape = (1, the_shape[1], the_shape[2], the_shape[3])
origReshaped = Reshape(shape)(attention_feature)
xReshaped = Reshape(shape)(high_feature)
together = Concatenate(axis=1)([origReshaped, xReshaped])
x = Unpooling()(together)
x = Conv2D(256, (1,1),padding='same',use_bias=False)(x)
x = pixel_shuffler((2, 2))(x)
# stage_six
x = Conv2D(64, (5, 5), activation='relu', padding='same', name='deconv1', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
x = BatchNormalization()(x)
#prediction layer
x = Conv2D(1, (5, 5), activation='sigmoid', padding='same', name='pred', kernel_initializer='he_normal',
bias_initializer='zeros')(x)
model = Model(inputs=resNet.input, outputs=x)
return model