Keras构建CNN模型时MaxPooling2D层出现Negative dimension报错如何解决

• 核心问题是特征图经过多层卷积池化后尺寸过小，无法满足第三个池化层的计算要求：
  在你使用padding='VALID'（默认值，无边缘填充）的配置下，逐层计算特征图宽高变化如下：
  1. 输入尺寸：28x28
  2. 第一层4x4卷积输出：25x25 → 2x2池化后输出12x12
  3. 第二层4x4卷积输出：9x9 → 2x2池化后输出4x4
  4. 第三层4x4卷积输出：1x1 → 此时输入第三个2x2池化层时，1x1的特征图小于池化核尺寸，计算时维度变为负数，触发报错。
• 隐藏问题：你在卷积层和全连接层之间没有添加Flatten()层，即使池化报错解决，后续运行也会因为维度不匹配报错。
• 冗余配置：只有Sequential的第一层需要指定input_shape，后续卷积层的input_shape参数不会生效，属于无用配置。

你可以选择以下任意一种方案修正代码：

方案1：删除第三个池化层，补全Flatten层（最简便） #

修正后代码如下：

from keras.models import Sequential
from keras.layers import Conv2D, MaxPool2D, Dense, Dropout, Flatten

model = Sequential()
model.add(Conv2D(filters = 32, kernel_size = (4,4), input_shape = (28,28,1), activation = 'relu'))
model.add(MaxPool2D(pool_size = (2,2)))

model.add(Conv2D(filters = 64, kernel_size = (4,4), activation = 'relu'))
model.add(MaxPool2D(pool_size=(2, 2)))

model.add(Conv2D(filters = 64, kernel_size = (4,4), activation = 'relu'))
# 移除第三个池化层

model.add(Flatten()) # 新增：将4维特征图展平为1维，适配全连接层输入
model.add(Dense(128, activation = 'relu'))
model.add(Dropout(0.5))

model.add(Dense(10, activation = 'softmax'))
model.compile(loss = 'categorical_crossentropy', optimizer = 'rmsprop', metrics = ['accuracy'])

方案2：调整卷积核尺寸，适配第三个池化层 #

将第三层卷积核改为3x3，第三层卷积输出变为2x2，可正常执行2x2池化：

from keras.models import Sequential
from keras.layers import Conv2D, MaxPool2D, Dense, Dropout, Flatten

model = Sequential()
model.add(Conv2D(filters = 32, kernel_size = (4,4), input_shape = (28,28,1), activation = 'relu'))
model.add(MaxPool2D(pool_size = (2,2)))

model.add(Conv2D(filters = 64, kernel_size = (4,4), activation = 'relu'))
model.add(MaxPool2D(pool_size=(2, 2)))

# 调整卷积核为3x3
model.add(Conv2D(filters = 64, kernel_size = (3,3), activation = 'relu'))
model.add(MaxPool2D(pool_size = (2,2)))

model.add(Flatten())
model.add(Dense(128, activation = 'relu'))
model.add(Dropout(0.5))

model.add(Dense(10, activation = 'softmax'))
model.compile(loss = 'categorical_crossentropy', optimizer = 'rmsprop', metrics = ['accuracy'])

方案3：给卷积层添加padding='same'配置 #

卷积时自动填充边缘，避免特征图尺寸快速缩小：

from keras.models import Sequential
from keras.layers import Conv2D, MaxPool2D, Dense, Dropout, Flatten

model = Sequential()
model.add(Conv2D(filters = 32, kernel_size = (4,4), padding='same', input_shape = (28,28,1), activation = 'relu'))
model.add(MaxPool2D(pool_size = (2,2)))

model.add(Conv2D(filters = 64, kernel_size = (4,4), padding='same', activation = 'relu'))
model.add(MaxPool2D(pool_size=(2, 2)))

model.add(Conv2D(filters = 64, kernel_size = (4,4), padding='same', activation = 'relu'))
model.add(MaxPool2D(pool_size = (2,2)))

model.add(Flatten())
model.add(Dense(128, activation = 'relu'))
model.add(Dropout(0.5))

model.add(Dense(10, activation = 'softmax'))
model.compile(loss = 'categorical_crossentropy', optimizer = 'rmsprop', metrics = ['accuracy'])

内容的提问来源于stack exchange，提问作者Tusar Ahmed