Keras InceptionV3 TypeError: unhashable type: 'Dimension'

Question

我正在尝试实现一个模型，它将灰度图像作为输入并返回一个数值作为输出。我使用 InceptionV3（从头开始训练）作为特征提取器，然后使用一些密集层进行最后阶段的回归。

这是我的代码：

from keras.applications.inception_v3 import InceptionV3
from keras.layers import Input, GlobalAveragePooling2D, Dense, Dropout, Flatten, BatchNormalization
from keras.models import Model
from keras.metrics import mean_absolute_error
from keras.utils import plot_model

inputs = Input(shape=(256, 256, 1))
x = BatchNormalization()(inputs)
x = InceptionV3(include_top = False, weights = None, input_shape=inputs.shape[1:])(x)
x = BatchNormalization()(x)
x = GlobalAveragePooling2D()(x)
x = Dense(1000, activation = 'relu' )(x)
x = Dense(1000, activation = 'relu' )(x)
outputs = Dense(1, activation = 'linear' )(x)
model = Model(inputs=inputs, outputs=outputs)

model.compile(optimizer = 'adam', loss = 'mse', metrics = [mae])

model.summary()

现在我在运行代码时收到此错误：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-36-50041eb640cc> in <module>()
      7 inputs = Input(shape=(256, 256, 1))
      8 x = BatchNormalization()(inputs)
----> 9 x = InceptionV3(include_top = False, weights = None, input_shape=inputs.shape[1:])(x)
     10 x = BatchNormalization()(x)
     11 x = GlobalAveragePooling2D()(x)

3 frames
/usr/local/lib/python3.6/dist-packages/keras_applications/imagenet_utils.py in _obtain_input_shape(input_shape, default_size, min_size, data_format, require_flatten, weights)
    273             default_shape = (input_shape[0], default_size, default_size)
    274         else:
--> 275             if input_shape[-1] not in {1, 3}:
    276                 warnings.warn(
    277                     'This model usually expects 1 or 3 input channels. '

TypeError: unhashable type: 'Dimension'

我不明白是什么导致了错误，因为当我使用顺序模型时它绝对没问题。但它不适用于此功能模型。

Answer 1

inputs.shape 不是列表，因此会引发错误。它为您提供tensorflow.python.framework.tensor_shape.TensorShape 类型的形状，其中包含Dimension 类型的每个维度的列表

print(inputs.shape)
# output TensorShape([Dimension(None), Dimension(256), Dimension(256), Dimension(1)])

您可以使用as_list() 获取形状作为列表：

# inputs.shape.as_list()
# output [None, 256, 256, 1]

x = InceptionV3(include_top = False, weights = None, input_shape=inputs.shape.as_list()[1:])(x)