Question

我有28x28张图片的数据集。数据点数组x的形状为(10000, 28, 28)，标签数组y的形状为(10000,)。
以下代码：

x = x.reshape(-1, 28, 28, 1)
model = Sequential([
    Conv2D(8, kernel_size=(3, 3), padding="same", activation=tf.nn.relu, input_shape=(28, 28, 1)),
    Dense(64, activation=tf.nn.relu),
    Dense(64, activation=tf.nn.relu),
    Dense(10, activation=tf.nn.softmax)
])
model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)
model.fit(x, y, epochs=5) #error

给予：

ValueError: Error when checking target: expected dense_3 to have 4 dimensions, but got array with shape (10000, 1)

model.summary()输出：

Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 28, 28, 8)         80        
_________________________________________________________________
dense_1 (Dense)              (None, 28, 28, 64)        576       
_________________________________________________________________
dense_2 (Dense)              (None, 28, 28, 64)        4160      
_________________________________________________________________
dense_3 (Dense)              (None, 28, 28, 10)        650       
=================================================================
Total params: 5,466
Trainable params: 5,466
Non-trainable params: 0
_________________________________________________________________

Answer 1

您忘记添加Flatten()层（keras.layers.Flatten()）：

model = Sequential([
    Conv2D(8, kernel_size=(3, 3), padding="same", activation=tf.nn.relu, input_shape=(28, 28, 1)),
    Flatten(),
    Dense(64, activation=tf.nn.relu),
    Dense(64, activation=tf.nn.relu),
    Dense(10, activation=tf.nn.softmax)
])

Answer 2

您的输出是3维的，而目标是1维的。您可能会在Flatten层之后缺少Con2D层，这会将卷积的输出减小到一个维度：

from keras.models import Sequential
from keras.layers import Conv2D, Dense, Flatten

# Fake data
import numpy as np
x = np.ones((10000, 28, 28))
y = np.ones((10000,))

x = x.reshape(-1, 28, 28, 1)
model = Sequential([
    Conv2D(8, kernel_size=(3, 3), padding="same", activation="relu", input_shape=(28, 28, 1)),
    Flatten(),
    Dense(64, activation="relu"),
    Dense(64, activation="relu"),
    Dense(10, activation="softmax")
])

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)

model.summary()
model.fit(x, y, epochs=1)

然后，尺寸正确：

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 28, 28, 8)         80        
_________________________________________________________________
flatten_1 (Flatten)          (None, 6272)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 64)                401472    
_________________________________________________________________
dense_2 (Dense)              (None, 64)                4160      
_________________________________________________________________
dense_3 (Dense)              (None, 10)                650       
=================================================================
Total params: 406,362
Trainable params: 406,362
Non-trainable params: 0

Keras提供错误-ValueError：检查目标时出错：预期density_3具有4维，但数组的形状为（10000，1）

2 个答案: