TF model ValueError: Shapes (None, 2,1)和(None, 4,4)不兼容

Question

我有训练和测试数据的形状如下:

(9, 28, 28, 1), (3, 28, 28, 1), (9, 2, 1), (3, 2, 1) 

的代码如下:

import tensorflow as tf
import keras
from keras.layers import Conv2D, MaxPooling2D
from keras.models import Sequential,Input,Model
from keras.layers import Dense, Dropout
batch_size = 28
epochs = 10
num_classes = 1
model = Sequential()
model.add(Conv2D(32, kernel_size=(3,3),padding='same', input_shape=(9,784,1), 
activation='linear'))
model.add(MaxPooling2D((2, 2),padding='same'))
model.add(Conv2D(64, (3, 3), activation='linear',padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
model.add(Conv2D(128, (3, 3), activation='linear',padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))

model.add(Dense(num_classes, activation='sigmoid'))

model.compile(loss=keras.losses.categorical_crossentropy, 
optimizer=keras.optimizers.Adam(),metrics=['accuracy'])
model.fit(X_train, y_train, 
batch_size=batch_size,epochs=epochs,verbose=1,validation_data=(X_test, y_test))

报错:

ValueError: Shapes (None, 2,1) and (None, 4,4)是不兼容的

Answer 1

除了你的数据形状和分配最终神经元without flattening的权重之外，你没有提到适当的input shape

这里我适当地添加了

import tensorflow as tf
from tensorflow import keras
from keras.layers import Conv2D, MaxPooling2D
from keras.models import Sequential,Input,Model
from keras.layers import Dense, Dropout, Flatten
batch_size = 28
epochs = 10
num_classes = 2
model = Sequential()
model.add(Conv2D(32, kernel_size=(3,3),padding='same', input_shape=(28,28,1), 
activation='linear'))
model.add(MaxPooling2D((2, 2),padding='same'))
model.add(Conv2D(64, (3, 3), activation='linear',padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
model.add(Conv2D(128, (3, 3), activation='linear',padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
model.add(Flatten())
model.add(Dense(12))
for layer in model.layers:
print(layer.get_output_at(0).get_shape().as_list())
model.add(Dense(2))
model.add(Dense(num_classes, activation='sigmoid'))

model.compile(loss=keras.losses.categorical_crossentropy, 
optimizer=keras.optimizers.Adam(),metrics=['accuracy'])
model.fit(X_train, y_train, batch_size=batch_size,epochs=epochs,verbose=1)#,validation_data=(X_test, y_test))

:

[None, 28, 28, 32]
[None, 14, 14, 32]
[None, 14, 14, 64]
[None, 7, 7, 64]
[None, 7, 7, 128]
[None, 4, 4, 128]
[None, 2048]
[None, 12]
Epoch 1/10
1/1 [==============================] - 7s 7s/step - loss: 0.7242 - accuracy: 0.5556
Epoch 2/10
1/1 [==============================] - 0s 6ms/step - loss: 0.7474 - accuracy: 0.4444
Epoch 3/10
1/1 [==============================] - 0s 6ms/step - loss: 0.7997 - accuracy: 0.4444
Epoch 4/10
1/1 [==============================] - 0s 6ms/step - loss: 0.8712 - accuracy: 0.4444
Epoch 5/10
1/1 [==============================] - 0s 5ms/step - loss: 0.9182 - accuracy: 0.4444
Epoch 6/10
1/1 [==============================] - 0s 5ms/step - loss: 0.9220 - accuracy: 0.4444
Epoch 7/10
1/1 [==============================] - 0s 5ms/step - loss: 0.9011 - accuracy: 0.4444
Epoch 8/10
1/1 [==============================] - 0s 5ms/step - loss: 0.8745 - accuracy: 0.4444
Epoch 9/10
1/1 [==============================] - 0s 5ms/step - loss: 0.8506 - accuracy: 0.4444
Epoch 10/10
1/1 [==============================] - 0s 7ms/step - loss: 0.8342 - accuracy: 0.4444
<keras.callbacks.History at 0x7fcf40808050>