ValueError：输入通道数与过滤器的相应维度不匹配，512！= 3答案

【问题标题】：ValueError: number of input channels does not match corresponding dimension of filter, 512 != 3ValueError：输入通道数与过滤器的相应维度不匹配，512！= 3
【发布时间】：2018-10-09 12:48:34
【问题描述】：

似乎similar problem，尝试了它的解决方案，但得到了

AttributeError: 'Tensor' object has no attribute 'reshape'

我正在使用keras建立一个基于VGG16的模型，如下代码所示

if K.image_data_format() == 'channels_first':
    input_shape = (3, 256, 256)
else:
    input_shape = (256, 256, 3)
input_image = Input(shape=input_shape)
base_model = VGG16( weights='imagenet', include_top=False, input_shape=input_shape)

这是基本模型输入输出

base_model.input
base_model.output
<tf.Tensor 'input_14:0' shape=(?, 256, 256, 3) dtype=float32>
<tf.Tensor 'block5_pool_13/MaxPool:0' shape=(?, 8, 8, 512) dtype=float32>

在这里我们可以看到，输入通道为3，但它返回输出通道为512。（不确定它是否与抛出的异常有关。）据我了解，它与 CNN 输入层中的通道不匹配。我没有热到修复它？

这是 CNN 层：

model = Sequential()
model.add(Conv2D(32, (5, 5), input_shape=input_shape))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(64, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Flatten())
model.add(Dense(64))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(3, activation='softmax'))
#model.add(Activation('softmax'))
model = Model(inputs=base_model.input, outputs=model(base_model.output))
model.compile(loss='categorical_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

Model() 方法抛出以下错误：

ValueError: number of input channels does not match corresponding dimension of filter, 512 != 3

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-66-bace1b0f7f30> in <module>()
     23 model.add(Dense(3, activation='softmax'))
     24 #model.add(Activation('softmax'))
---> 25 model = Model(inputs=base_model.input, outputs=model(base_model.output))
     26 model.compile(loss='categorical_crossentropy',
     27               optimizer='rmsprop',

/usr/local/lib/python3.6/dist-packages/keras/engine/topology.py in __call__(self, inputs, **kwargs)
    617 
    618             # Actually call the layer, collecting output(s), mask(s), and shape(s).
--> 619             output = self.call(inputs, **kwargs)
    620             output_mask = self.compute_mask(inputs, previous_mask)
    621 

/usr/local/lib/python3.6/dist-packages/keras/models.py in call(self, inputs, mask)
    577         if not self.built:
    578             self.build()
--> 579         return self.model.call(inputs, mask)
    580 
    581     def build(self, input_shape=None):

/usr/local/lib/python3.6/dist-packages/keras/engine/topology.py in call(self, inputs, mask)
   2083             return self._output_tensor_cache[cache_key]
   2084         else:
-> 2085             output_tensors, _, _ = self.run_internal_graph(inputs, masks)
   2086             return output_tensors
   2087 

/usr/local/lib/python3.6/dist-packages/keras/engine/topology.py in run_internal_graph(self, inputs, masks)
   2233                                 if 'mask' not in kwargs:
   2234                                     kwargs['mask'] = computed_mask
-> 2235                             output_tensors = _to_list(layer.call(computed_tensor, **kwargs))
   2236                             output_masks = layer.compute_mask(computed_tensor,
   2237                                                               computed_mask)

/usr/local/lib/python3.6/dist-packages/keras/layers/convolutional.py in call(self, inputs)
    166                 padding=self.padding,
    167                 data_format=self.data_format,
--> 168                 dilation_rate=self.dilation_rate)
    169         if self.rank == 3:
    170             outputs = K.conv3d(

/usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py in conv2d(x, kernel, strides, padding, data_format, dilation_rate)
   3339         strides=strides,
   3340         padding=padding,
-> 3341         data_format=tf_data_format)
   3342 
   3343     if data_format == 'channels_first' and tf_data_format == 'NHWC':

/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/nn_ops.py in convolution(input, filter, padding, strides, dilation_rate, name, data_format)
    779         dilation_rate=dilation_rate,
    780         name=name,
--> 781         data_format=data_format)
    782     return op(input, filter)
    783 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/nn_ops.py in __init__(self, input_shape, filter_shape, padding, strides, dilation_rate, name, data_format)
    839           "number of input channels does not match corresponding dimension of "
    840           "filter, {} != {}".format(input_channels_dim,
--> 841                                     filter_shape[num_spatial_dims]))
    842 
    843     strides, dilation_rate = _get_strides_and_dilation_rate(

ValueError: number of input channels does not match corresponding dimension of filter, 512 != 3

【问题讨论】：

您可以尝试只修改您编写的图层的输入
即将model.add(Reshape(target_shape=(128, 128, 2), input_shape=list(base_model.output.get_shape().as_list()[1:]))) 添加为模型的第一层。
@Primusa - 谢谢，似乎错误已经消失。我理解它重塑了维度。但没有得到这条线的作用？更具体地说，这里的 target_shape 是什么？
目标形状是形状转换成的样子。它需要一个 input_shape 并将其转换为 target_shape
转换成的形状的大小和输入的形状必须相同。 8*8*512 == 128*128*2，所以你可以将输出从VGG16转换为128、128、2。只要乘积相同，实际数字或多或少是任意的。

标签： python keras-layer channels

【解决方案1】：

由于您无法对 VGG16 的输出做任何事情，我认为您可以只修改输入层：

将此添加为模型的第一层：

model.add(Reshape(target_shape=(128, 128, 2), input_shape=list(base_model.output.get_shape().as_list()[1:])))

重塑层的作用是接收input_shape，然后将形状更改为target_shape。只要输入和目标的大小都是恒定的（所有数字的乘积相同），就允许进行此操作。

【讨论】：