称为 fc3 的全连接层如何具有 20x125 的权重答案

【问题标题】：how does fully connected layer called fc3 have 20x125 weights称为 fc3 的全连接层如何具有 20x125 的权重
【发布时间】：2017-08-27 12:06:09
【问题描述】：

根据我的计算，池化输出应该是 5x4x4（5 个大小为 4x4 的特征图），因此将其展平将产生一个 1x80 的向量。因此 fc3 应该具有 20x80 的权重，但 pycaffe 显示的层具有 20x125 的权重。这是原始文本文件。这是我的计算

等式是使用 (dimension_size - kernel)/stride + 1

转换 1：1x5x26x26
池 1：1x5x12x12
转换 2：1x5x10x10
池2：1x5x4x4

input: "data"
input_shape {
  dim: 1
  dim: 1
  dim: 28
  dim: 28
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 5
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 5
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.1
    }
  }
}
layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc3"
  type: "InnerProduct"
  bottom: "pool2"
  top: "fc3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 20
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 0.1
    }
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "fc3"
  top: "fc3"
}
layer {
  name: "drop3"
  type: "Dropout"
  bottom: "fc3"
  top: "fc3"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc4"
  type: "InnerProduct"
  bottom: "fc3"
  top: "fc4"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 10
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "softmax"
  type: "Softmax"
  bottom: "fc4"
  top: "softmax"
}

【问题讨论】：

标签： caffe

【解决方案1】：

事实证明，对于最大池化，caffe 会向上取整输出大小，而不是截断如果它是分数。因此它会产生不正确的形状。该问题已在 GitHub 上报告，但由于向后兼容性问题尚未解决。

【讨论】：