使用预训练Keras模型时遭遇维度形状错误

问题背景 #

需要基于带预训练权重的VGG16模型，追加自定义简易CNN结构，参考Keras教程改造后，训练时出现MaxPooling2D负维度错误，希望保留MaxPooling层以对比模型结果，寻求快速修复方案。

原自定义CNN代码 #

model = Sequential()
model.add(Rescaling(1.0 / 255))
model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(256,256,3)))
model.add(MaxPool2D(pool_size=(2, 2), strides=2))
model.add(Conv2D(64, kernel_size=(3, 3), activation='relu'))
model.add(MaxPool2D(pool_size=(2, 2), strides=2))
model.add(Flatten())
model.add(Dense(units=5, activation='softmax'))

改造后代码 #

x = base_model.output
x = Rescaling(1.0 / 255)(x)
x = Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(256,256,3))(x)
x = MaxPool2D(pool_size=(2, 2), strides=2)(x)
x = Conv2D(64, kernel_size=(3, 3), activation='relu')(x)
x = MaxPool2D(pool_size=(2, 2), strides=2)(x)
x = GlobalAveragePooling2D()(x)
predictions = Dense(units=5, activation='softmax')(x)

错误信息 #

ValueError: Exception encountered when calling layer "max_pooling2d_7" (type MaxPooling2D).

Negative dimension size caused by subtracting 2 from 1 for '{{node model/max_pooling2d_7/MaxPool}} = MaxPool[T=DT_FLOAT, data_format="NHWC", explicit_paddings=[], ksize=[1, 2, 2, 1], padding="VALID", strides=[1, 2, 2, 1]](model/conv2d_10/Relu)' with input shapes: [?,1,1,64].

Call arguments received:
  • inputs=tf.Tensor(shape=(None, 1, 1, 64), dtype=float32)

快速修复方案 #

错误本质是VGG16输出的特征图经过自定义CNN的卷积、池化后，尺寸缩小到1x1，此时再执行2x2的MaxPooling就会出现负维度。以下是几种保留MaxPooling层的修复方法：

方法1：给卷积层添加padding='same' #

让卷积操作不缩小特征图尺寸，仅通过池化层降低维度，避免过早出现1x1的特征图：

# 初始化VGG16时指定输入形状（匹配你的256x256输入）
base_model = VGG16(weights='imagenet', include_top=False, input_shape=(256,256,3))

x = base_model.output
# 移除多余的Rescaling：VGG16预训练权重对应输入无需额外归一化，改用preprocess_input处理数据
x = Conv2D(32, kernel_size=(3, 3), activation='relu', padding='same')(x)  # 添加padding='same'，去掉无用的input_shape
x = MaxPool2D(pool_size=(2, 2), strides=2)(x)
x = Conv2D(64, kernel_size=(3, 3), activation='relu', padding='same')(x)  # 添加padding='same'
x = MaxPool2D(pool_size=(2, 2), strides=2)(x)
x = GlobalAveragePooling2D()(x)
predictions = Dense(units=5, activation='softmax')(x)

方法2：给MaxPooling2D添加padding='same' #

即使特征图尺寸不足2x2，通过填充保证池化后维度为正：

x = base_model.output
x = Conv2D(32, kernel_size=(3, 3), activation='relu')(x)
x = MaxPool2D(pool_size=(2, 2), strides=2, padding='same')(x)  # 添加padding='same'
x = Conv2D(64, kernel_size=(3, 3), activation='relu')(x)
x = MaxPool2D(pool_size=(2, 2), strides=2, padding='same')(x)  # 添加padding='same'
x = GlobalAveragePooling2D()(x)
predictions = Dense(units=5, activation='softmax')(x)

额外注意事项 #

• 初始化VGG16时必须指定input_shape=(256,256,3)且include_top=False，这样输出的特征图尺寸为8x8x512，足够支撑后续的卷积池化操作；
• 移除Rescaling(1.0/255)，改用tf.keras.applications.vgg16.preprocess_input处理输入数据，匹配VGG16预训练权重的输入要求。

内容的提问来源于stack exchange，提问作者brosefzai