尝试在 keras 中向 CNN model 添加输入层

Question

I tried to add input to a parallel path cnn, to make a residual architecture, but I am getting dimension mismatch.

from keras import layers, Model
input_shape = (128,128,3) # Change this accordingly
my_input = layers.Input(shape=input_shape) # one input
def parallel_layers(my_input, parallel_id=1):
  x = layers.SeparableConv2D(32, (9, 9), activation='relu', name='conv_1_'+str(parallel_id))(my_input)
  x = layers.MaxPooling2D(2, 2)(x)
  x = layers.SeparableConv2D(64, (9, 9), activation='relu', name='conv_2_'+str(parallel_id))(x)
  x = layers.MaxPooling2D(2, 2)(x)
  x = layers.SeparableConv2D(128, (9, 9), activation='relu', name='conv_3_'+str(parallel_id))(x)
  x = layers.MaxPooling2D(2, 2)(x)
  x = layers.Flatten()(x)
  x = layers.Dropout(0.5)(x)
  x = layers.Dense(512, activation='relu')(x)

  return x

parallel1 = parallel_layers(my_input, 1)
parallel2 = parallel_layers(my_input, 2)

concat = layers.Concatenate()([parallel1, parallel2])
concat=layers.Add()(concat,my_input)
x = layers.Dense(128, activation='relu')(concat)
x = Dense(7, activation='softmax')(x)

final_model = Model(inputs=my_input, outputs=x)

final_model.fit_generator(train_generator, steps_per_epoch = 
    nb_train_samples // batch_size, epochs = epochs, validation_data = validation_generator,
    validation_steps = nb_validation_samples // batch_size) 

I am getting the error

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-48-163442df0d4c> in <module>()
      1 concat = layers.Concatenate()([parallel1, parallel2])
----> 2 concat=layers.Add()(concat,my_input)
      3 x = layers.Dense(128, activation='relu')(parallel2)
      4 x = Dense(7, activation='softmax')(x)
      5 

TypeError: __call__() takes 2 positional arguments but 3 were given

I am using keras 2.1.6 version. Kindly help to resolve this final_model.summary()

Answer 1

define your add layer in this way

concat=layers.Add()([concat,my_input])

Answer 2

You have to remove the following line:

concat=layers.Add()(concat,my_input)

It does not make any sense. You have a method that takes an input, branches into two parallel models. The outputs of both of them ( parallel1 and parallel2 )are vectors of length 512 . Then you can either Concatenate them to have a length of 1024 or Add them to have a length of 512 again. The concat then goes through further Dense layers.

So in short, remove the following line:

concat=layers.Add()(concat,my_input)

If you want to concatenate and have a vector of length 1024, keep the rest of the code as it is, otherwise, if you want to add them and have a vector of length 512 instead, replace the following line:

concat = layers.Concatenate()([parallel1, parallel2])

with this:

concat = layers.Add()([parallel1, parallel2])