neural-network - caffe 丢失错误 : Check failed: The data and label should have the same first dimension

Question

这是我的 Blob 形状和图层:

-------------------------------- Blob

data                        4096     4.10e+03    (1, 2, 1, 2048)
Convolution1               32736     3.27e+04    (1, 16, 1, 2046)
ReLU1                      32736     3.27e+04    (1, 16, 1, 2046)
Convolution2               32704     3.27e+04    (1, 16, 1, 2044)
ReLU2                      32704     3.27e+04    (1, 16, 1, 2044)
...
Crop4                       4224     4.22e+03    (1, 16, 1, 264)
Concat4                     8448     8.45e+03    (1, 32, 1, 264)
Convolution17               4192     4.19e+03    (1, 16, 1, 262)
ReLU21                      4192     4.19e+03    (1, 16, 1, 262)
Convolution18               4160     4.16e+03    (1, 16, 1, 260)
unet1                       4160     4.16e+03    (1, 16, 1, 260)
ampl0                       4096     4.10e+03    (1, 4096)
Reshape0                    4096     4.10e+03    (1, 1, 1, 4096)
conv1                      65472     6.55e+04    (1, 16, 1, 4092)
conv1_conv1_0_split_0      65472     6.55e+04    (1, 16, 1, 4092)
conv1_conv1_0_split_1      65472     6.55e+04    (1, 16, 1, 4092)
Scale1                     65472     6.55e+04    (1, 16, 1, 4092)
ReLU22                     65472     6.55e+04    (1, 16, 1, 4092)
Scale2                     65472     6.55e+04    (1, 16, 1, 4092)
...
ReLU28                    517120     5.17e+05    (1, 128, 8, 505)
Scale8                    517120     5.17e+05    (1, 128, 8, 505)
ReLU29                    517120     5.17e+05    (1, 128, 8, 505)
crelu4                   1034240     1.03e+06    (1, 128, 16, 505)
maxPool4                  518144     5.18e+05    (1, 128, 16, 253)
ampl                          21     2.10e+01    (1, 21)

我在损失层得到的错误:

F0416 15:43:21.957676 95620 loss_layer.cpp:19] Check failed: bottom[0]->shape(0) == bottom[1]->shape(0) (1 vs. 10) The data and label should have the same first dimension.

注意:这个错误是我在CNN层中间加了一个全连接层(ampl0)+ reshape (Reshape0)层后出现的。没有它们，它工作得很好!

感谢您的帮助。更新:那些完全连接和 reshape 层是:

layer {
  name: "ampl0"
  type: "InnerProduct"
  bottom: "unet1"
  top: "ampl0"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  inner_product_param {
    num_output: 4096
    bias_term: false
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.2
    }
  }
}
layer {
  name: "Reshape0"
  type: "Reshape"
  bottom: "ampl0"
  top: "Reshape0"
  reshape_param {
    shape {
      dim: 1
      dim: 1
      dim: 1
      dim:-1
    }
  }
}

Answer 1

你的 "Reshape"层强制第一个维度 (batch_size) 为 1，因此当您更改 batch_size 时，您的净中断。
为避免这种情况，您需要“Reshape” 来复制第一个维度:

  layer {
    name: "reshape"
    type: "Reshape"
    bottom: "input"
    top: "output"
    reshape_param {
      shape {
        dim: 0  # copy the dimension from below  <-- !!
        dim: 1  # insert singleton dimension
        dim: 1
        dim: -1 # infer it from the other dimensions
      }
    }
  }

我想

    reshape_param { shape { dim: 1 dim: 1 }  num_axes: 0 axis: 1 }

也可能为您带来诡计。

有关“Reshape” 参数的更多信息和选项，请参阅caffe.proto :

  // axis and num_axes control the portion of the bottom blob's shape that are
  // replaced by (included in) the reshape. By default (axis == 0 and
  // num_axes == -1), the entire bottom blob shape is included in the reshape,
  // and hence the shape field must specify the entire output shape.
  //
  // axis may be non-zero to retain some portion of the beginning of the input
  // shape (and may be negative to index from the end; e.g., -1 to begin the
  // reshape after the last axis, including nothing in the reshape,
  // -2 to include only the last axis, etc.).
  //
  // For example, suppose "input" is a 2D blob with shape 2 x 8.
  // Then the following ReshapeLayer specifications are all equivalent,
  // producing a blob "output" with shape 2 x 2 x 4:
  //
  //   reshape_param { shape { dim: 2  dim: 2  dim: 4 } }
  //   reshape_param { shape { dim: 2  dim: 4 } axis:  1 }
  //   reshape_param { shape { dim: 2  dim: 4 } axis: -3 }
  //
  // num_axes specifies the extent of the reshape.
  // If num_axes >= 0 (and axis >= 0), the reshape will be performed only on
  // input axes in the range [axis, axis+num_axes].
  // num_axes may also be -1, the default, to include all remaining axes
  // (starting from axis).
  //
  // For example, suppose "input" is a 2D blob with shape 2 x 8.
  // Then the following ReshapeLayer specifications are equivalent,
  // producing a blob "output" with shape 1 x 2 x 8.
  //
  //   reshape_param { shape { dim:  1  dim: 2  dim:  8 } }
  //   reshape_param { shape { dim:  1  dim: 2  }  num_axes: 1 }
  //   reshape_param { shape { dim:  1  }  num_axes: 0 }
  //
  // On the other hand, these would produce output blob shape 2 x 1 x 8:
  //
  //   reshape_param { shape { dim: 2  dim: 1  dim: 8  }  }
  //   reshape_param { shape { dim: 1 }  axis: 1  num_axes: 0 }
  //