def build(input_shape, num_outputs):
print('original input shape:', input_shape)
_handle_dim_ordering()
if len(input_shape) != 4:
raise Exception("Input shape should be a tuple (nb_channels, kernel_dim1, kernel_dim2, kernel_dim3)")
print('original input shape:', input_shape)
if K.image_dim_ordering() == 'tf':
input_shape = (input_shape[1], input_shape[2], input_shape[3], input_shape[0])
print('change input shape:', input_shape)
input = Input(shape=input_shape)
# input = applyPCA(input, 30)
conv1 = Conv3D(filters=16, kernel_size=(3, 3, 7), strides=(1, 1, 5), kernel_regularizer=regularizers.l2(0.01),
kernel_initializer='he_normal', use_bias=False, activation='relu')(input)
conv1 = BatchNormalization()(conv1)
conv2 = Conv3D(filters=32, kernel_size=(3, 3, 5), strides=(1, 1, 3), kernel_regularizer=regularizers.l2(0.01),
kernel_initializer='he_normal', use_bias=False, activation='relu')(conv1)
conv2 = BatchNormalization()(conv2)
print(conv2.shape)
conv3 = Conv3D(filters=64, kernel_size=(3, 3, 3), strides=(1, 1, 1), kernel_regularizer=regularizers.l2(0.01),
kernel_initializer='he_normal', use_bias=False, activation='relu')(conv2)
conv3 = BatchNormalization()(conv3)
print(conv3.shape)
conv3 = SeparableConv3D(filters=128, kernel_size=(3, 3, 1), strides=(1, 1, 1),
kernel_initializer=regularizers.l2(0.01),
use_bias=False, activation='relu')(conv3)
print(conv3._keras_shape)
conv3_shape = conv3._keras_shape
l = Reshape((conv3_shape[1], conv3_shape[2], conv3_shape[3] * conv3_shape[4]))(conv3)
print(l)
# conv11
l = Conv2D(32, (1, 1), padding='same', kernel_regularizer=regularizers.l2(0.01), kernel_initializer='he_normal',
use_bias=False, activation='relu')(l)
l = BatchNormalization()(l)
print(l.shape)
# l = Conv2D(64, (1, 1), padding='same', kernel_regularizer=regularizers.l2(0.01), kernel_initializer='he_normal',
# use_bias=False, activation='relu')(l)
# l3 = BatchNormalization()(l)
# conv12
l_offset = ConvOffset2D(32)(l)
l1 = Conv2D(64, (3, 3), padding='same', strides=(1, 1), kernel_regularizer=regularizers.l2(0.01),
kernel_initializer='he_normal', use_bias=False, activation='relu')(l_offset)
l1 = BatchNormalization()(l1)
print(l1.shape)
# conv21
l_offset = ConvOffset2D(32)(l)
l2 = Conv2D(64, (5, 5), padding='same', strides=(1, 1), dilation_rate=5,
kernel_regularizer=regularizers.l2(0.01), kernel_initializer='he_normal', use_bias=False,
activation='relu')(l_offset)
l2 = BatchNormalization()(l2)
print(l2.shape)
l = keras.layers.add([l1, l2])
se = squeeze_excite_block(l)
l1 = multiply([l1, se])
l2 = multiply([l2, se])
l = keras.layers.add([l1, l2])
# out
l = GlobalAvgPool2D()(l)
# 输入分类器
# Classifier block
dense = Dense(units=num_outputs, activation="softmax", kernel_initializer="he_normal")(l)
model = Model(inputs=input, outputs=dense)
return model
def build(input_shape, num_outputs):
print('original input shape:', input_shape)
_handle_dim_ordering()
if len(input_shape) != 4:
raise Exception(
"Input shape should be a tuple (nb_channels, kernel_dim1, kernel_dim2, kernel_dim3)"
)
print('original input shape:', input_shape)
# orignal input shape: 1,7,7,200
if K.image_dim_ordering() == 'tf':
input_shape = (input_shape[1], input_shape[2], input_shape[3],
input_shape[0])
print('change input shape:', input_shape)
# 用keras中函数式模型API，不用序贯模型API
# 张量流输入
input = Input(shape=input_shape)
# print(input.shape)
# ( ?,7,7,200,1 )
conv1 = Conv3D(filters=16,
kernel_size=(11, 11, 24),
strides=(12, 12, 5),
kernel_regularizer=regularizers.l2(0.01))(input)
print(conv1.shape)
# ( None, 1,1,16,16 )
l = Reshape((16, 16, 1))(conv1)
print(l)
# conv11
l = Conv2D(32, (3, 3), padding='same', name='conv11',
trainable=False)(l)
l = Activation('relu', name='conv11_relu')(l)
l = BatchNormalization(name='conv11_bn')(l)
# conv12
l_offset = ConvOffset2D(32, name='conv12_offset')(l)
l = Conv2D(64, (3, 3),
padding='same',
strides=(2, 2),
name='conv12',
trainable=False)(l_offset)
l = Activation('relu', name='conv12_relu')(l)
l = BatchNormalization(name='conv12_bn')(l)
# conv21
l_offset = ConvOffset2D(64, name='conv21_offset')(l)
l = Conv2D(64, (3, 3),
padding='same',
strides=(2, 2),
name='conv21',
trainable=False)(l_offset)
l = Activation('relu', name='conv21_relu')(l)
l = BatchNormalization(name='conv21_bn')(l)
# out
l = GlobalAvgPool2D(name='avg_pool')(l)
# 输入分类器
# Classifier block
dense = Dense(units=num_outputs,
activation="softmax",
kernel_initializer="he_normal")(l)
model = Model(inputs=input, outputs=dense)
return model