经典网络结构：resnet18结构

版权声明：本文为博主原创文章，转载请注明出处，谢谢。 https://blog.csdn.net/wfei101/article/details/83050491

Resnet 网络结构：

layer {
  name: "res2c"
  type: "Eltwise"
  bottom: "res2b"
  bottom: "res2c_branch2c"
  top: "res2c"
}
layer {
  name: "res2c_relu"
  type: "ReLU"
  bottom: "res2c"
  top: "res2c"
}
layer {
  name: "res3a_branch1"
  type: "Convolution"
  bottom: "res2c"
  top: "res3a_branch1"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "bn3a_branch1"
  type: "BatchNorm"
  bottom: "res3a_branch1"
  top: "res3a_branch1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "scale3a_branch1"
  type: "Scale"
  bottom: "res3a_branch1"
  top: "res3a_branch1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "res3a_branch2a"
  type: "Convolution"
  bottom: "res2c"
  top: "res3a_branch2a"
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 0
    kernel_size: 1
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "bn3a_branch2a"
  type: "BatchNorm"
  bottom: "res3a_branch2a"
  top: "res3a_branch2a"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "scale3a_branch2a"
  type: "Scale"
  bottom: "res3a_branch2a"
  top: "res3a_branch2a"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "res3a_branch2a_relu"
  type: "ReLU"
  bottom: "res3a_branch2a"
  top: "res3a_branch2a"
}
layer {
  name: "res3a_branch2b"
  type: "Convolution"
  bottom: "res3a_branch2a"
  top: "res3a_branch2b"
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "bn3a_branch2b"
  type: "BatchNorm"
  bottom: "res3a_branch2b"
  top: "res3a_branch2b"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "scale3a_branch2b"
  type: "Scale"
  bottom: "res3a_branch2b"
  top: "res3a_branch2b"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "res3a_branch2b_relu"
  type: "ReLU"
  bottom: "res3a_branch2b"
  top: "res3a_branch2b"
}
layer {
  name: "res3a_branch2c"
  type: "Convolution"
  bottom: "res3a_branch2b"
  top: "res3a_branch2c"
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "bn3a_branch2c"
  type: "BatchNorm"
  bottom: "res3a_branch2c"
  top: "res3a_branch2c"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "scale3a_branch2c"
  type: "Scale"
  bottom: "res3a_branch2c"
  top: "res3a_branch2c"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "res3a"
  type: "Eltwise"
  bottom: "res3a_branch1"
  bottom: "res3a_branch2c"
  top: "res3a"
}
layer {
  name: "res3a_relu"
  type: "ReLU"
  bottom: "res3a"
  top: "res3a"
}

proto结构：

如果是SSD的最后第八层，则出现卷积是3*3的卷积pading =0；步长是1的卷积，最后生成1*1的卷积核；384的输入的话！

网址： https://github.com/BigcowPeking/resnet-imagenet-caffe