import os
from keras import layers
from keras.layers import Input, merge
from keras.layers.convolutional import (AveragePooling2D, Conv2D, MaxPooling2D,
ZeroPadding2D)
from keras.layers.core import Activation, Dense, Flatten
from keras.layers.normalization import BatchNormalization
from keras.models import Model,Sequential
from keras.utils import plot_model
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
def identity_block(x, nb_filter, kernel_size=3):
k1, k2, k3 = nb_filter
shortcut = x
out = Conv2D(k1, kernel_size=(1,1), strides=(1,1),padding="valid",activation="relu")(x)
out = BatchNormalization(axis=3)(out)
out = Conv2D(k2, kernel_size=(3,3), strides=(1,1), padding='same',activation="relu")(out)
out = BatchNormalization(axis=3)(out)
out = Conv2D(k3, kernel_size=(1,1), strides=(1,1),padding="valid")(out)
out = BatchNormalization(axis=3)(out)
# out = merge([out, shortcut], mode='sum')
out= layers.add([out,shortcut])
out = Activation('relu')(out)
return out
def conv_block(x, nb_filter, kernel_size=3):
k1, k2, k3 = nb_filter
shortcut = x
out = Conv2D(k1, kernel_size=(1,1), strides=(2,2), padding="valid",activation="relu")(x)
out = BatchNormalization(axis=3)(out)
out = out = Conv2D(k2, kernel_size=(kernel_size,kernel_size), strides=(1,1), padding="same",activation="relu")(out)
out = BatchNormalization()(out)
out = Conv2D(k3, kernel_size=(1,1), strides=(1,1), padding="valid")(out)
out = BatchNormalization(axis=3)(out)
shortcut = Conv2D(k3, kernel_size=(1,1), strides=(2,2), padding="valid")(shortcut)
shortcut = BatchNormalization(axis=3)(shortcut)
# out = merge([out, shortcut], mode='sum')
out = layers.add([out, shortcut])
out = Activation('relu')(out)
return out
def buildNet():
inp = Input(shape=(224, 224, 3))
out = ZeroPadding2D((3, 3))(inp)
out = Conv2D(64, kernel_size=(7, 7), strides=(2, 2),activation="relu")(out)
out = BatchNormalization()(out)
out = MaxPooling2D(pool_size=(3, 3), strides=(2, 2), padding="same")(out)
out = conv_block(out, [64, 64, 256])
out = identity_block(out, [64, 64, 256])
out = identity_block(out, [64, 64, 256])
out = conv_block(out, [128, 128, 512])
out = identity_block(out, [128, 128, 512])
out = identity_block(out, [128, 128, 512])
out = identity_block(out, [128, 128, 512])
out = conv_block(out, [256, 256, 1024])
out = identity_block(out, [256, 256, 1024])
out = identity_block(out, [256, 256, 1024])
out = identity_block(out, [256, 256, 1024])
out = identity_block(out, [256, 256, 1024])
out = identity_block(out, [256, 256, 1024])
out = conv_block(out, [512, 512, 2048])
out = identity_block(out, [512, 512, 2048])
out = identity_block(out, [512, 512, 2048])
out = AveragePooling2D((4, 4))(out)
out = Flatten()(out) # 展平
out = Dense(1000, activation='softmax')(out)
model = Model(inputs=inp, outputs=out)
return model
if __name__ == '__main__':
# resNet18 = ResNet(block_num=[2,2,2,2])
# resNet34 = ResNet(block_num=[3,4,6,3])
# resNet50 = ResNet(block_num=[3,4,6,3])
# resNet101 = ResNet(block_num=[3,4,23,3])
# resNet152 = ResNet(block_num=[3,8,36,3])
net = buildNet()
net.compile(optimizer='adam',loss='categorical_crossentropy', metrics=['accuracy'])
plot_model(net, to_file='./models/resnet.png')
net.summary()
