Keras功能性api多输入模型期望1个输入数组，但得到2个。我该怎么做，以便我的模型期望2个数组作为输入？

Question

  

ValueError：检查模型目标时出错：Numpy数组列表   您传递给模型的信息不是模型期望的大小。   预计将看到1个数组，但得到以下2个列表   数组。

import numpy as np import scipy import matplotlib

from sklearn.model_selection import train_test_split

import tensorflow as tf import keras, keras.layers as L

from keras.layers.merge import concatenate



from skimage import io import glob x=glob.glob('/home/tschaub/img_input/img_sets/118_18333/*.png') y=[io.imread(i) for i in x] import numpy as np len(y) z=np.stack(y, axis=0)

z=np.reshape(z,[len(y),384,384,1]) X=z img_shape=(384,384,1) code_size=32 from keras.models import Model from keras.layers import Input


t=glob.glob('/home/tschaub/img_input/img_sets/118_18333/*.png') j=[io.imread(i) for i in t[20:]]

len(j) s=np.stack(j, axis=0)

s=np.reshape(s,[len(j),384,384,1]) Z=s


#------------------------------------------------------------------------------------------------------
#encoder_1 input_tensor_1 = L.Input(shape=(384,384,1),name='Input_1')

conv1_1=L.Conv2D(32, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv1_1')(input_tensor_1)

mp1_1=L.MaxPool2D(pool_size=(2, 2), name='mp1_1')(conv1_1)

conv1_2=L.Conv2D(64, kernel_size=(3, 3),strides=1, padding='same', activation='elu',name='conv1_2')(mp1_1)

mp1_2=L.MaxPool2D(pool_size=(2, 2), name='mp1_2')(conv1_2)

conv1_3=L.Conv2D(128, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv1_3')(mp1_2)

mp1_3=L.MaxPool2D(pool_size=(2, 2), name='mp1_3')(conv1_3)

conv1_4=L.Conv2D(256, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv1_4')(mp1_3)

mp1_4=L.MaxPool2D(pool_size=(2, 2), name='mp1_4')(conv1_4)

vector1=L.Flatten(name='vector1')(mp1_4)

target_vector1=L.Dense(code_size, name='target_1')(vector1)
#encoder_2 input_tensor_2 = L.Input(shape=(384,384,1),name='Input_2')

conv2_1=L.Conv2D(32, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv2_1')(input_tensor_2)

mp2_1=L.MaxPool2D(pool_size=(2, 2), name='mp2_1')(conv2_1)

conv2_2=L.Conv2D(64, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv2_2')(mp2_1)

mp2_2=L.MaxPool2D(pool_size=(2, 2), name='mp2_2')(conv2_2)

conv2_3=L.Conv2D(128, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv2_3')(mp2_2)

mp2_3=L.MaxPool2D(pool_size=(2, 2), name='mp2_3')(conv2_3)

conv2_4=L.Conv2D(256, kernel_size=(3, 3),strides=1, padding='same', activation='elu', name='conv2_4')(mp2_3)

mp2_4=L.MaxPool2D(pool_size=(2, 2), name='mp2_4')(conv2_4)

vector2=L.Flatten(name='vector2')(mp2_4)

target_vector2=L.Dense(code_size, name='target_2')(vector2)


#merge code from encoders merge=concatenate([target_vector1, target_vector2])


# decoder

d_input_1=L.Dense(64, name='de_in')(merge)

dense_1=L.Dense(147456, name='de_dense1')(d_input_1)

reshape_1=L.Reshape((24,24,256), name='reshape')(dense_1)

conTR1_1=L.Conv2DTranspose(128, kernel_size=(3, 3),strides=2, padding='same', activation='elu', name='up_1')(reshape_1)

conTR1_2=L.Conv2DTranspose(filters=64, kernel_size=(3, 3), strides=2, activation='elu', padding='same', name='up_2')(conTR1_1)

conTR1_3=L.Conv2DTranspose(filters=32, kernel_size=(3, 3), strides=2, activation='elu', padding='same', name='up_3')(conTR1_2)

conTR1_4=L.Conv2DTranspose(filters=1, kernel_size=(3, 3), strides=2, activation=None, padding='same', name='up_4')(conTR1_3)



X_train, X_test = train_test_split(X, test_size=0.1, random_state=42) Z_train, Z_test = train_test_split(Z, test_size=0.1, random_state=42)

from keras.utils import multi_gpu_model 
autoencoder = Model(inputs=[input_tensor_1, input_tensor_2],outputs=[conTR1_4]) 
autoencoder.summary()
parallel_model=multi_gpu_model(autoencoder,gpus=4)

parallel_model.compile('adamax','mse')

from keras.utils import plot_model plot_model(autoencoder, to_file='time.png')

parallel_model.fit([X_train,Z_train], [X_train,Z_train], epochs=120, batch_size=10)

我认为我指定了两个输入：

 autoencoder = Model(inputs=[input_tensor_1, input_tensor_2],outputs=[conTR1_4])

并适合两个输入：

  parallel_model.fit([X_train,Z_train], [X_train,Z_train], epochs=120, batch_size=10)

但是由于某些原因，我仍然收到错误消息，表明我的模型只需要一个输入。在实例化模型和拟合过程中，我尝试了各种语法，但似乎无法弄清楚发生了什么！

请帮助！ （提前感谢）