时间戳

首页 > 解决方案 > 如何解决“ValueError：“连接”层需要具有匹配形状的输入，连接轴除外”？

python - 如何解决“ValueError：“连接”层需要具有匹配形状的输入，连接轴除外”？

问题描述

我正在尝试实现 3D CNN，但由于输入形状不匹配，我收到一条值错误消息。我做错了什么来得到这个错误，我该如何解决这种问题？

traceback 中引用的部分代码如下：

    x = conv3d_bn(img_input, 32, 3, 3, 3, strides=(2, 2, 2), padding='same', name='Conv3d_1b_3x3')
    x = conv3d_bn(img_input, 64, 3, 3, 3, strides=(2, 2, 2), padding='same', name='Conv3d_1c_3x3')

    branch_0 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), padding='same', name='MaxPool2d_2a_3x3')(x)

    branch_1 = conv3d_bn(branch_0, 96, 3, 3, 3, padding='same', name='Conv3d_2b_3x3')

    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_2a')

    branch_0 = conv3d_bn(x, 64, 1, 1, 1, padding='same', name='Conv3d_3_0a_1x1')
    branch_0 = conv3d_bn(branch_0, 96, 3, 3, 3, padding='same', name='Conv3d_3c_0b_3x3')

    branch_1 = conv3d_bn(x, 64, 1, 1, 1, padding='same', name='Conv3d_3_1a_1x1')
    branch_1 = conv3d_bn(branch_1, 64, 7, 1, 1, padding='same', name='Conv3d_3c_1b_3x3')
    branch_1 = conv3d_bn(branch_1, 64, 1, 7, 7, padding='same', name='Conv3d_3c_1c_3x3')
    branch_1 = conv3d_bn(branch_1, 96, 3, 3, 3, padding='same', name='Conv3d_3c_1d_3x3')

    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_3a')

    branch_0 = conv3d_bn(x, 192, 1, 1, 1, padding='same', name='Conv3d_3_0a_1x1')

    branch_1 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), padding='same', name='MaxPool2d_0b_3x3')(x)

    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_4a')x = conv3d_bn(img_input, 32, 3, 3, 3, strides=(2, 2, 2), padding='same', name='Conv3d_1b_3x3')
    x = conv3d_bn(img_input, 64, 3, 3, 3, strides=(2, 2, 2), padding='same', name='Conv3d_1c_3x3')

    branch_0 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), padding='same', name='MaxPool2d_2a_3x3')(x)

    branch_1 = conv3d_bn(branch_0, 96, 3, 3, 3, padding='same', name='Conv3d_2b_3x3')

    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_2a')

    branch_0 = conv3d_bn(x, 64, 1, 1, 1, padding='same', name='Conv3d_3_0a_1x1')
    branch_0 = conv3d_bn(branch_0, 96, 3, 3, 3, padding='same', name='Conv3d_3c_0b_3x3')

    branch_1 = conv3d_bn(x, 64, 1, 1, 1, padding='same', name='Conv3d_3_1a_1x1')
    branch_1 = conv3d_bn(branch_1, 64, 7, 1, 1, padding='same', name='Conv3d_3c_1b_3x3')
    branch_1 = conv3d_bn(branch_1, 64, 1, 7, 7, padding='same', name='Conv3d_3c_1c_3x3')
    branch_1 = conv3d_bn(branch_1, 96, 3, 3, 3, padding='same', name='Conv3d_3c_1d_3x3')

    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_3a')

    branch_0 = conv3d_bn(x, 192, 1, 1, 1, padding='same', name='Conv3d_3_0a_1x1')

    branch_1 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), padding='same', name='MaxPool2d_0b_3x3')(x)

    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_4a')

这是回溯：

Traceback (most recent call last):
  File "train.py", line 292, in <module>
    main(**vars(p.parse_args()))
  File "train.py", line 155, in main
    400, spatial_squeeze=True, endpoint_logit='Logits')
  File "/home/larry/Documents/Projekt/i3dv2.py", line 241, in InceptionI3DV2
    x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_4a')
  File "/home/larry/anaconda3/lib/python3.7/site-packages/keras/layers/merge.py", line 649, in concatenate
    return Concatenate(axis=axis, **kwargs)(inputs)
  File "/home/larry/anaconda3/lib/python3.7/site-packages/keras/engine/base_layer.py", line 463, in __call__
    self.build(unpack_singleton(input_shapes))
  File "/home/larry/anaconda3/lib/python3.7/site-packages/keras/layers/merge.py", line 362, in build
    'Got inputs shapes: %s' % (input_shape))
ValueError: A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 32, 75, 75, 192), (None, 32, 38, 38, 192)]

我试过这个

 branch_0 = conv3d_bn(x, 192, 1, 3, 3, padding='same', name='Conv3d_3_0a_1x1')
 branch_1 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), padding='same', name='MaxPool2d_0b_3x3')(x) 
 x = layers.concatenate([branch_0, branch_1], axis=channel_axis, name='mixed_4a')

仍然得到同样的错误。有人可以准确解释我需要做什么来解决这个错误。谢谢。

期待您的评论。

标签： pythontensorflowkerasconcatenation

解决方案

您的错误指出：

  • branch0形状是(None, 32, 75, 75, 192)
  • branch1形状是(None, 32, 38, 38, 192)

这是您定义层的方式（branch0 保持大小，而 branch1 是池化的）：

branch_0 = conv3d_bn(x, 192, 1, 3, 3, ...)
branch_1 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), ...)(x)

从您的定义来看Conv3d_2b_3x3：

branch_0 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), ...)(x)
branch_1 = conv3d_bn(branch_0, 96, 3, 3, 3, ... name='Conv3d_2b_3x3')

我假设你的意思是：

branch_0 = MaxPooling3D((1, 3, 3), strides=(1, 2, 2), padding='same', name='MaxPool2d_0b_3x3')(x)
branch_1 = conv3d_bn(branch_0, 192, 1, 1, 1, padding='same', name='Conv3d_3_0a_1x1')

推荐阅读