def build(self, _subname=None):
f = self.conf.filters
name = self.conf.name + _subname if _subname else self.conf.name
g_input = Input(shape=self.conf.input_shape)
# (batch_size, 208, 160, 1)
conv1_1 = conv2D_layer_bn(g_input, name=None, filters=32, kernel_size=7, strides=1, padding='same',
activation='relu')
conv1_1 = conv2D_layer_bn(conv1_1, name=None, filters=32, kernel_size=7, strides=1, padding='same',
activation='relu')
# (batch_size, 208, 160, 32)
conv2_1 = g_downsampling_bn(conv1_1, filters=f*2)
# (batch_size, 104, 80, 64)
conv3_1 = g_downsampling_bn(conv2_1, filters=f*4)
# (batch_size, 52, 40, 128)
# ========== residual part ====================================
res_1_1 = residual_block_bn(conv3_1, f*4)
# (batch_size, 52, 40, 128)
res_1_2 = residual_block_bn(res_1_1, f*4)
# (batch_size, 52, 40, 128)
res_1_3 = residual_block_bn(res_1_2, f*4)
# (batch_size, 52, 40, 128)
res_1_4 = residual_block_bn(res_1_3, f*4)
# (batch_size, 52, 40, 128)
res_1_5 = residual_block_bn(res_1_4, f*4)
# (batch_size, 52, 40, 128)
res_1_6 = residual_block_bn(res_1_5, f*4)
# (batch_size, 52, 40, 128)
# ========== residual part end ================================
conv4_1 = g_upsampling_bn(res_1_6, f*4)
# (batch_size, 104, 80, 128)
concat_1 = Concatenate()([conv4_1, conv2_1])
# (batch_size, 104, 80, 192)
conv4_2 = conv2D_layer_bn(concat_1, name="conv4_2", filters=f*2, kernel_size=3, strides=1, padding='same',
activation='relu')
# (batch_size, 104, 80, 64)
conv5_1 = g_upsampling_bn(conv4_2, f*2)
# (batch_size, 208, 160, 64)
concat_2 = Concatenate()([conv5_1, conv1_1])
# (batch_size, 208, 160, 96)
conv5_2 = conv2D_layer_bn(concat_2, name="conv5_1", filters=f, kernel_size=3, strides=1, padding='same',
activation='relu')
# (batch_size, 208, 160, 32)
conv5_3 = Conv2D(filters=1, name="conv5_3", kernel_size=7, strides=1, padding="same", activation="relu")(conv5_2)
# (batch_size, 208, 160, 1)
g_output = Activation(self.conf.G_activation)(conv5_3)
# (batch_size, 208, 160, 1)
self.model = Model(inputs=g_input, outputs=g_output, name=name)
self.model.summary()
def build(self, _subname=None):
f = self.conf.filters
name = self.conf.name + _subname if _subname else self.conf.name
g_input = Input(shape=self.conf.input_shape)
# (batch size, 160, 208, 1)
conv1_1 = conv2D_layer_bn(g_input,
name="conv1_1",
filters=f,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv1_2 = conv2D_layer_bn(conv1_1,
name="conv1_2",
filters=f,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
pool1 = MaxPool2D()(conv1_2) # (batch size, 80, 104, filters)
# (batch size, 80, 104, filters)
conv2_1 = conv2D_layer_bn(pool1,
name="conv2_1",
filters=f * 2,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv2_2 = conv2D_layer_bn(conv2_1,
name="conv2_2",
filters=f * 2,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
pool2 = MaxPool2D()(conv2_2) # (batch size, 40, 52, filters*2)
# (batch size, 40, 52, filters*2)
conv3_1 = conv2D_layer_bn(pool2,
name="conv3_1",
filters=f * 4,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv3_2 = conv2D_layer_bn(conv3_1,
name="conv3_2",
filters=f * 4,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
pool3 = MaxPool2D()(conv3_2) # (batch size, 20, 26, filters*4)
# (batch size, 20, 26, filters*4)
conv4_1 = conv2D_layer_bn(pool3,
name="conv4_1",
filters=f * 8,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv4_2 = conv2D_layer_bn(conv4_1,
name="conv4_2",
filters=f * 8,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
pool4 = MaxPool2D()(conv4_2)
# (batch size, 10, 13, filters*8)
# ======================================================================================================================
# In middle layer, we concatenate Age vector
# (batch size, 10, 13, filters*8)
mid1_1 = conv2D_layer_bn(pool4,
name="mid1_1",
filters=f,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 10, 13, filter)
flat1_1 = Flatten()(mid1_1)
# (batch size, 4160)
dens1_1 = Dense(units=self.conf.latent_space,
name="dens1_1",
activation='sigmoid')(flat1_1)
dens1_1 = BatchNormalization()(dens1_1)
age_vector = Input(
shape=(self.conf.age_dim, )) # Age vector shape: (20,)
# (batch size, 130)
mid_concat1_1 = Concatenate()(
[dens1_1, age_vector]) # Encode age vector into middle layer
# (batch size, 130+100)
dens2_1 = Dense(units=4160, name="dens2_1",
activation='relu')(mid_concat1_1)
# (batch size, 4160)
rshape1_1 = Reshape(target_shape=(10, 13, 32))(dens2_1)
# (batch size, 10, 13, 32)
mid_concat2_1 = Concatenate()([pool4, rshape1_1])
# (batch size, 10, 13, 32+f*8)
# ======================================================================================================================
upconv4 = deconv2D_layer_bn(mid_concat2_1,
name="upconv4",
filters=f * 8,
kernel_size=3,
strides=2,
padding='same',
activation='relu',
kernel_initializer="he_normal")
concat4 = Concatenate()([upconv4, conv4_2])
conv_m_1_1 = conv2D_layer_bn(concat4,
name='conv_m_1_1',
filters=f * 8,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv_m_1_2 = conv2D_layer_bn(conv_m_1_1,
name='conv_m_1_2',
filters=f * 8,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 20, 26, filters*8)
upconv3 = deconv2D_layer_bn(conv_m_1_2,
name="upconv3",
filters=f * 4,
kernel_size=3,
strides=2,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 40, 52, filters*4)
concat3 = Concatenate()([upconv3, conv3_2])
# (batch size, 40, 52, filters*8)
conv5_1 = conv2D_layer_bn(concat3,
name='conv5_1',
filters=f * 4,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv5_2 = conv2D_layer_bn(conv5_1,
name='conv5_2',
filters=f * 4,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 40, 52, filters*4)
upconv2 = deconv2D_layer_bn(conv5_2,
name="upconv2",
filters=f * 2,
kernel_size=3,
strides=2,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 80, 104, filters*2)
concat2 = Concatenate()([upconv2, conv2_2])
# (batch size, 80, 104, filters*4)
conv6_1 = conv2D_layer_bn(concat2,
name='conv6_1',
filters=f * 2,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv6_2 = conv2D_layer_bn(conv6_1,
name='conv6_2',
filters=f * 2,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 80, 104, filters*2)
upconv1 = deconv2D_layer_bn(conv6_2,
name="upconv1",
filters=f,
kernel_size=3,
strides=2,
padding='same',
activation='relu',
kernel_initializer="he_normal")
# (batch size, 160, 208, filters)
concat1 = Concatenate()([upconv1, conv1_2])
# (batch size, 160, 208, filters*2)
conv8_1 = conv2D_layer_bn(concat1,
name="conv8_1",
filters=f,
kernel_size=3,
strides=1,
padding='same',
activation='relu',
kernel_initializer="he_normal")
conv8_2 = conv2D_layer_bn(conv8_1,
name="conv8_2",
filters=1,
kernel_size=3,
strides=1,
padding='same',
activation=self.conf.G_activation,
kernel_initializer="he_normal")
# (batch size, 160, 208, 1)
self.model = Model(inputs=[g_input, age_vector],
outputs=conv8_2,
name=name)
self.model.summary()
def build(self, _subname = None):
inp_shape = self.conf.input_shape
f = self.conf.filters
name = self.conf.name+_subname if _subname else self.conf.name
d_input = Input(inp_shape)
# (batch size, 160, 208, 1)
conv1_1 = conv2D_layer(d_input, name="conv1_1", filters=f, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
pool1 = MaxPool2D()(conv1_1) # (batch size, 80, 104, filters)
# (batch size, 80, 104, filters)
conv2_1 = conv2D_layer(pool1, name="conv2_1", filters=f*2, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
pool2 = MaxPool2D()(conv2_1) # (batch size, 40, 52, filters*2)
# (batch size, 40, 52, filters*2)
conv3_1 = conv2D_layer(pool2, name="conv3_1", filters=f*4, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
conv3_2 = conv2D_layer(conv3_1, name="conv3_2", filters=f*4, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
pool3 = MaxPool2D()(conv3_2) # (batch size, 20, 26, filters*4)
# (batch size, 20, 26, filters*4)
conv4_1 = conv2D_layer(pool3, name="conv4_1", filters=f*8, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
conv4_2 = conv2D_layer(conv4_1, name="conv4_2", filters=f*8, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
pool4 = MaxPool2D()(conv4_2)
# (batch size, 10, 13, filters*8)
# =============================================================================================================
# In middle layer, we concatenate Age vector
# (batch size, 10, 13, filters*8)
mid1_1 = conv2D_layer_bn(pool4, name="mid1_1", filters=f, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal")
# (batch size, 10, 13, filter)
flat1_1 = Flatten()(mid1_1)
# (batch size, 4160)
dens1_1 = Dense(units=130, name="dens1_1", activation='sigmoid')(flat1_1)
dens1_1 = BatchNormalization()(dens1_1)
age_vector = Input(shape=(self.conf.age_dim,)) # Age vector shape: (20,)
# (batch size, 130)
mid_concat1_1 = Concatenate()([dens1_1, age_vector]) # Encode age vector into middle layer
# (batch size, 130+20)
dens2_1 = Dense(units=130, name="dens2_1", activation='relu')(mid_concat1_1)
dens2_1 = BatchNormalization()(dens2_1)
AD_vector = Input(shape=(self.conf.AD_dim,))
mid_concat2_1 = Concatenate()([dens2_1, AD_vector])
dens3_1 = Dense(units=4160, name="dens3_1", activation='relu')(mid_concat2_1)
# (batch size, 4160)
rshape1_1 = Reshape(target_shape=(10, 13, 32))(dens3_1)
# (batch size, 10, 13, 32)
mid_concat3_1 = Concatenate()([pool4, rshape1_1])
# (batch size, 10, 13, 32+f*8)
# =============================================================================================================
conv5_1 = conv2D_layer(mid_concat3_1, name="conv5_1", filters=f*16, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal" )
conv5_2 = conv2D_layer(conv5_1, name="conv5_2", filters=f*16, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal" )
# (batch size, 10, 13, filters*16)
convD_1 = conv2D_layer(conv5_2, name="convD_1", filters=f*16, kernel_size=3, strides=1, padding='same',
activation='relu', kernel_initializer="he_normal" )
convD_2 = conv2D_layer(convD_1, name="convD_2", filters=1, kernel_size=3, strides=1, padding='same',
activation='linear', kernel_initializer="he_normal")
# (batch size, 10, 13, filters*16)
averagePool = GlobalAveragePooling2D()(convD_2)