Keras共享某个层

对一个层的多次调用，就是在共享这个层。

input1 = Input(shape=[28,28])
input2 = Input(shape=[28,28])
x1 = Flatten()(input1)
x1 = Dense(60,activation="relu")(x1)
x2 = Flatten()(input2)
x2 = Dense(60,activation="relu")(x2)

d = Dense(10, activation='softmax')

output1 = d(x1)
output2 = d(x2)
model1 = Model(input=[input1], output=[output1])
model2 = Model(input=[input1], output=[output1])
print(model1.summary())
print(model2.summary())

结果：

____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
input_1 (InputLayer)             (None, 28, 28)        0                                            
____________________________________________________________________________________________________
flatten_1 (Flatten)              (None, 784)           0           input_1[0][0]                    
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 60)            47100       flatten_1[0][0]                  
____________________________________________________________________________________________________
dense_3 (Dense)                  (None, 10)            610         dense_1[0][0]                    
====================================================================================================
Total params: 47710
____________________________________________________________________________________________________
None
____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
input_1 (InputLayer)             (None, 28, 28)        0                                            
____________________________________________________________________________________________________
flatten_1 (Flatten)              (None, 784)           0           input_1[0][0]                    
____________________________________________________________________________________________________
dense_1 (Dense)                  (None, 60)            47100       flatten_1[0][0]                  
____________________________________________________________________________________________________
dense_3 (Dense)                  (None, 10)            610         dense_1[0][0]                    
====================================================================================================
Total params: 47710
____________________________________________________________________________________________________
None

如果修改x2：

x2 = Dense(70,activation="relu")(x2)

就会报错：

Exception: Input 0 is incompatible with layer dense_3: expected shape=(None, 60), found shape=(None, 70)

就是因为第一次使用Dense层的时候，他的参数w的大小已经定了是（28*28，60），共享层的目的不就是为了共享参数的吗，所以，我讲明白了吧！