介绍
CNN中有权重共享的概念，是一张图片中的多个区域共享同一个卷积核的参数。与前者不同，这里讨论的是权重共享网络，即有多个不同的输入，经过具有相同参数的一个网络，得到两个不同的结果。
有一类网络叫孪生网络（如MatchNet）就具有权重共享的特性。这种网络可以用来衡量两个样本的相似性，用于判定两个样本是不是同一个类。在人脸识别、签名识别、语言识别任务中有一些应用。
keras实现孪生网络的方式很简单，只要将需要共享的地方组成一个Model。在后续网络调用Model即可。最后整个网络再组成一个Model。相当于整体Model包含共享的Model。。。Model套Model
keras官方文档示例
下面ClassFilerNet1是共享权重的网络，最后summary的参数量是4万多；ClassFilerNet2没有共享权重，summary的参数量是9万多。
共享的部分是conv2d-conv2d-maxpooling2d-flatten
第一个网络将共享的部分组成了vision_model，后续像Layer一样函数式调用即可。
第二个网络重新创建了不同的Layer，没有组成一个Model，所以参数会增加。
import keras
from keras.layers import Conv2D, MaxPooling2D, Input, Dense, Flatten
from keras.models import Model
def ClassiFilerNet1():
# First, define the vision modules
digit_input = Input(shape=(27, 27, 1))
x = Conv2D(64, (3, 3))(digit_input)
x = Conv2D(64, (3, 3))(x)
x = MaxPooling2D((2, 2))(x)
out = Flatten()(x)
vision_model = Model(digit_input, out)
# Then define the tell-digits-apart model
digit_a = Input(shape=(27, 27, 1))
digit_b = Input(shape=(27, 27, 1))
# The vision model will be shared, weights and all
out_a = vision_model(digit_a)
out_b = vision_model(digit_b)
concatenated = keras.layers.concatenate([out_a, out_b])
out = Dense(1, activation='sigmoid')(concatenated)
classification_model = Model([digit_a, digit_b], out)
return classification_model
def ClassiFilerNet2():
digit_a = Input(shape=(27, 27, 1))
x = Conv2D(64, (3, 3))(digit_a)
x = Conv2D(64, (3, 3))(x)
x = MaxPooling2D((2, 2))(x)
out1 = Flatten()(x)
digit_b = Input(shape=(27, 27, 1))
x = Conv2D(64, (3, 3))(digit_b)
x = Conv2D(64, (3, 3))(x)
x = MaxPooling2D((2, 2))(x)
out2 = Flatten()(x)
concatenated = keras.layers.concatenate([out1, out2])
out = Dense(1, activation='sigmoid')(concatenated)
classification_model = Model([digit_a, digit_b], out)
return classification_model
MatchNet 出自这里
代码是直接粘贴过来的，实际上核心用法看keras官方文档最简单也最直观。粘贴过来只是做个记录。
MatchNet包含两块：FeatureExtract，Classification。
FeatureExtract部分，对于多个输入应该采用相同的权重进行处理，需要用到共享权重。
Classification对提取的特征进行分类，判别是不是同一类。
下面第一段代码是非共享权重，参数量为4.8M；第二段是共享权重，参数量为3.4M。
可以放到tensorboard查看网络结构，有一些差异。
from keras.models import Sequential
from keras.layers import merge, Conv2D, MaxPool2D, Activation, Dense, concatenate, Flatten
from keras.layers import Input
from keras.models import Model
from keras.utils import np_utils
import tensorflow as tf
import keras
from keras.datasets import mnist
import numpy as np
from keras.utils import np_utils
from keras.callbacks import EarlyStopping, ModelCheckpoint, TensorBoard, ReduceLROnPlateau
from keras.utils.vis_utils import plot_model
# ---------------------函数功能区-------------------------
def FeatureNetwork():
"""生成特征提取网络"""
"""这是根据，MNIST数据调整的网络结构，下面注释掉的部分是，原始的Matchnet网络中feature network结构"""
inp = Input(shape = (28, 28, 1), name='FeatureNet_ImageInput')
models = Conv2D(filters=24, kernel_size=(3, 3), strides=1, padding='same')(inp)
models = Activation('relu')(models)
models = MaxPool2D(pool_size=(3, 3))(models)
models = Conv2D(filters=64, kernel_size=(3, 3), strides=1, padding='same')(models)
# models = MaxPool2D(pool_size=(3, 3), strides=(2, 2))(models)
models = Activation('relu')(models)
models = Conv2D(filters=96, kernel_size=(3, 3), strides=1, padding='valid')(models)
models = Activation('relu')(models)
models = Conv2D(filters=96, kernel_size=(3, 3), strides=1, padding='valid')(models)
models = Activation('relu')(models)
models = Flatten()(models)
models = Dense(512)(models)
models = Activation('relu')(models)
model = Model(inputs=inp, outputs=models)
return model
def ClassiFilerNet(): # add classifier Net
"""生成度量网络和决策网络，其实maychnet是两个网络结构，一个是特征提取层(孪生)，一个度量层+匹配层(统称为决策层)"""
input1 = FeatureNetwork() # 孪生网络中的一个特征提取
input2 = FeatureNetwork() # 孪生网络中的另一个特征提取
for layer in input2.layers: # 这个for循环一定要加，否则网络重名会出错。
layer.name = layer.name + str("_2")
inp1 = input1.input
inp2 = input2.input
merge_layers = concatenate([input1.output, input2.output]) # 进行融合，使用的是默认的sum，即简单的相加
fc1 = Dense(1024, activation='relu')(merge_layers)
fc2 = Dense(1024, activation='relu')(fc1)
fc3 = Dense(2, activation='softmax')(fc2)
class_models = Model(inputs=[inp1, inp2], outputs=[fc3])
print("1111")
return class_models
from keras.models import Sequential
from keras.layers import merge, Conv2D, MaxPool2D, Activation, Dense, concatenate, Flatten
from keras.layers import Input
from keras.models import Model
from keras.utils import np_utils
import tensorflow as tf
import keras
from keras.datasets import mnist
import numpy as np
from keras.utils import np_utils
from keras.callbacks import EarlyStopping, ModelCheckpoint, TensorBoard, ReduceLROnPlateau
from keras.utils.vis_utils import plot_model
# ----------------函数功能区-----------------------
def FeatureNetwork():
"""生成特征提取网络"""
"""这是根据，MNIST数据调整的网络结构，下面注释掉的部分是，原始的Matchnet网络中feature network结构"""
inp = Input(shape = (28, 28, 1), name='FeatureNet_ImageInput')
models = Conv2D(filters=24, kernel_size=(3, 3), strides=1, padding='same')(inp)
models = Activation('relu')(models)
models = MaxPool2D(pool_size=(3, 3))(models)
models = Conv2D(filters=64, kernel_size=(3, 3), strides=1, padding='same')(models)
# models = MaxPool2D(pool_size=(3, 3), strides=(2, 2))(models)
models = Activation('relu')(models)
models = Conv2D(filters=96, kernel_size=(3, 3), strides=1, padding='valid')(models)
models = Activation('relu')(models)
models = Conv2D(filters=96, kernel_size=(3, 3), strides=1, padding='valid')(models)
models = Activation('relu')(models)
# models = Conv2D(64, kernel_size=(3, 3), strides=2, padding='valid')(models)
# models = Activation('relu')(models)
# models = MaxPool2D(pool_size=(3, 3), strides=(2, 2))(models)
models = Flatten()(models)
models = Dense(512)(models)
models = Activation('relu')(models)
model = Model(inputs=inp, outputs=models)
return model
def ClassiFilerNet(reuse=True): # add classifier Net
"""生成度量网络和决策网络，其实maychnet是两个网络结构，一个是特征提取层(孪生)，一个度量层+匹配层(统称为决策层)"""
if reuse:
inp = Input(shape=(28, 28, 1), name='FeatureNet_ImageInput')
models = Conv2D(filters=24, kernel_size=(3, 3), strides=1, padding='same')(inp)
models = Activation('relu')(models)
models = MaxPool2D(pool_size=(3, 3))(models)
models = Conv2D(filters=64, kernel_size=(3, 3), strides=1, padding='same')(models)
# models = MaxPool2D(pool_size=(3, 3), strides=(2, 2))(models)
models = Activation('relu')(models)
models = Conv2D(filters=96, kernel_size=(3, 3), strides=1, padding='valid')(models)
models = Activation('relu')(models)
models = Conv2D(filters=96, kernel_size=(3, 3), strides=1, padding='valid')(models)
models = Activation('relu')(models)
# models = Conv2D(64, kernel_size=(3, 3), strides=2, padding='valid')(models)
# models = Activation('relu')(models)
# models = MaxPool2D(pool_size=(3, 3), strides=(2, 2))(models)
models = Flatten()(models)
models = Dense(512)(models)
models = Activation('relu')(models)
model = Model(inputs=inp, outputs=models)
inp1 = Input(shape=(28, 28, 1)) # 创建输入
inp2 = Input(shape=(28, 28, 1)) # 创建输入2
model_1 = model(inp1) # 孪生网络中的一个特征提取分支
model_2 = model(inp2) # 孪生网络中的另一个特征提取分支
merge_layers = concatenate([model_1, model_2]) # 进行融合，使用的是默认的sum，即简单的相加
else:
input1 = FeatureNetwork() # 孪生网络中的一个特征提取
input2 = FeatureNetwork() # 孪生网络中的另一个特征提取
for layer in input2.layers: # 这个for循环一定要加，否则网络重名会出错。
layer.name = layer.name + str("_2")
inp1 = input1.input
inp2 = input2.input
merge_layers = concatenate([input1.output, input2.output]) # 进行融合，使用的是默认的sum，即简单的相加
fc1 = Dense(1024, activation='relu')(merge_layers)
fc2 = Dense(1024, activation='relu')(fc1)
fc3 = Dense(2, activation='softmax')(fc2)
class_models = Model(inputs=[inp1, inp2], outputs=[fc3])
print("22222")
return class_models