deploy_MnasNet.prototxt

input: "data"
input_dim: 1
input_dim: 3
input_dim: 224
input_dim: 224
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  convolution_param {
    num_output: 32
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv1/bn"
  type: "BatchNorm"
  bottom: "conv1"
  top: "conv1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv1/scale"
  type: "Scale"
  bottom: "conv1"
  top: "conv1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv1/ReLU"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "conv2_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv1"
  top: "conv2_1/dw"
  convolution_param {
    num_output: 32
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv2_1/dw"
  top: "conv2_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv2_1/dw/scale"
  type: "Scale"
  bottom: "conv2_1/dw"
  top: "conv2_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv2_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv2_1/dw"
  top: "conv2_1/dw"
}
layer {
  name: "conv2_1/sep"
  type: "Convolution"
  bottom: "conv2_1/dw"
  top: "conv2_1/sep"
  convolution_param {
    num_output: 16
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_1/sep/bn"
  type: "BatchNorm"
  bottom: "conv2_1/sep"
  top: "conv2_1/sep"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv2_1/sep/scale"
  type: "Scale"
  bottom: "conv2_1/sep"
  top: "conv2_1/sep"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv2_1/sep/ReLU"
  type: "ReLU"
  bottom: "conv2_1/sep"
  top: "conv2_1/sep"
}
layer {
  name: "conv3_1/sep1"
  type: "Convolution"
  bottom: "conv2_1/sep"
  top: "conv3_1/sep1"
  convolution_param {
    num_output: 48
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/sep1/bn"
  type: "BatchNorm"
  bottom: "conv3_1/sep1"
  top: "conv3_1/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_1/sep1/scale"
  type: "Scale"
  bottom: "conv3_1/sep1"
  top: "conv3_1/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_1/sep1/ReLU"
  type: "ReLU"
  bottom: "conv3_1/sep1"
  top: "conv3_1/sep1"
}
layer {
  name: "conv3_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv3_1/sep1"
  top: "conv3_1/dw"
  convolution_param {
    num_output: 48
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv3_1/dw"
  top: "conv3_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_1/dw/scale"
  type: "Scale"
  bottom: "conv3_1/dw"
  top: "conv3_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv3_1/dw"
  top: "conv3_1/dw"
}
layer {
  name: "conv3_1/sep2"
  type: "Convolution"
  bottom: "conv3_1/dw"
  top: "conv3_1/sep2"
  convolution_param {
    num_output: 24
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/sep2/bn"
  type: "BatchNorm"
  bottom: "conv3_1/sep2"
  top: "conv3_1/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_1/sep2/scale"
  type: "Scale"
  bottom: "conv3_1/sep2"
  top: "conv3_1/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_1/sep2/ReLU"
  type: "ReLU"
  bottom: "conv3_1/sep2"
  top: "conv3_1/sep2"
}
layer {
  name: "conv3_2/sep1"
  type: "Convolution"
  bottom: "conv3_1/sep2"
  top: "conv3_2/sep1"
  convolution_param {
    num_output: 72
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_2/sep1/bn"
  type: "BatchNorm"
  bottom: "conv3_2/sep1"
  top: "conv3_2/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_2/sep1/scale"
  type: "Scale"
  bottom: "conv3_2/sep1"
  top: "conv3_2/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_2/sep1/ReLU"
  type: "ReLU"
  bottom: "conv3_2/sep1"
  top: "conv3_2/sep1"
}
layer {
  name: "conv3_2/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv3_2/sep1"
  top: "conv3_2/dw"
  convolution_param {
    num_output: 72
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv3_2/dw"
  top: "conv3_2/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_2/dw/scale"
  type: "Scale"
  bottom: "conv3_2/dw"
  top: "conv3_2/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_2/dw/ReLU"
  type: "ReLU"
  bottom: "conv3_2/dw"
  top: "conv3_2/dw"
}
layer {
  name: "conv3_2/sep2"
  type: "Convolution"
  bottom: "conv3_2/dw"
  top: "conv3_2/sep2"
  convolution_param {
    num_output: 24
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_2/sep2/bn"
  type: "BatchNorm"
  bottom: "conv3_2/sep2"
  top: "conv3_2/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_2/sep2/scale"
  type: "Scale"
  bottom: "conv3_2/sep2"
  top: "conv3_2/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_2/Eltwise1"
  type: "Eltwise"
  bottom: "conv3_1/sep2"
  bottom: "conv3_2/sep2"
  top: "conv3_2/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv3_2/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv3_2/Eltwise1"
  top: "conv3_2/Eltwise1"
}
layer {
  name: "conv3_3/sep1"
  type: "Convolution"
  bottom: "conv3_2/Eltwise1"
  top: "conv3_3/sep1"
  convolution_param {
    num_output: 72
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_3/sep1/bn"
  type: "BatchNorm"
  bottom: "conv3_3/sep1"
  top: "conv3_3/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_3/sep1/scale"
  type: "Scale"
  bottom: "conv3_3/sep1"
  top: "conv3_3/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_3/sep1/ReLU"
  type: "ReLU"
  bottom: "conv3_3/sep1"
  top: "conv3_3/sep1"
}
layer {
  name: "conv3_3/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv3_3/sep1"
  top: "conv3_3/dw"
  convolution_param {
    num_output: 72
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_3/dw/bn"
  type: "BatchNorm"
  bottom: "conv3_3/dw"
  top: "conv3_3/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_3/dw/scale"
  type: "Scale"
  bottom: "conv3_3/dw"
  top: "conv3_3/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_3/dw/ReLU"
  type: "ReLU"
  bottom: "conv3_3/dw"
  top: "conv3_3/dw"
}
layer {
  name: "conv3_3/sep2"
  type: "Convolution"
  bottom: "conv3_3/dw"
  top: "conv3_3/sep2"
  convolution_param {
    num_output: 24
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_3/sep2/bn"
  type: "BatchNorm"
  bottom: "conv3_3/sep2"
  top: "conv3_3/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv3_3/sep2/scale"
  type: "Scale"
  bottom: "conv3_3/sep2"
  top: "conv3_3/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_3/Eltwise1"
  type: "Eltwise"
  bottom: "conv3_2/Eltwise1"
  bottom: "conv3_3/sep2"
  top: "conv3_3/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv3_3/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv3_3/Eltwise1"
  top: "conv3_3/Eltwise1"
}
layer {
  name: "conv4_1/sep1"
  type: "Convolution"
  bottom: "conv3_3/Eltwise1"
  top: "conv4_1/sep1"
  convolution_param {
    num_output: 72
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/sep1/bn"
  type: "BatchNorm"
  bottom: "conv4_1/sep1"
  top: "conv4_1/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_1/sep1/scale"
  type: "Scale"
  bottom: "conv4_1/sep1"
  top: "conv4_1/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_1/sep1/ReLU"
  type: "ReLU"
  bottom: "conv4_1/sep1"
  top: "conv4_1/sep1"
}
layer {
  name: "conv4_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv4_1/sep1"
  top: "conv4_1/dw"
  convolution_param {
    num_output: 72
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv4_1/dw"
  top: "conv4_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_1/dw/scale"
  type: "Scale"
  bottom: "conv4_1/dw"
  top: "conv4_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv4_1/dw"
  top: "conv4_1/dw"
}
layer {
  name: "conv4_1/sep2"
  type: "Convolution"
  bottom: "conv4_1/dw"
  top: "conv4_1/sep2"
  convolution_param {
    num_output: 40
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/sep2/bn"
  type: "BatchNorm"
  bottom: "conv4_1/sep2"
  top: "conv4_1/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_1/sep2/scale"
  type: "Scale"
  bottom: "conv4_1/sep2"
  top: "conv4_1/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_1/sep2/ReLU"
  type: "ReLU"
  bottom: "conv4_1/sep2"
  top: "conv4_1/sep2"
}
layer {
  name: "conv4_2/sep1"
  type: "Convolution"
  bottom: "conv4_1/sep2"
  top: "conv4_2/sep1"
  convolution_param {
    num_output: 120
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_2/sep1/bn"
  type: "BatchNorm"
  bottom: "conv4_2/sep1"
  top: "conv4_2/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_2/sep1/scale"
  type: "Scale"
  bottom: "conv4_2/sep1"
  top: "conv4_2/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_2/sep1/ReLU"
  type: "ReLU"
  bottom: "conv4_2/sep1"
  top: "conv4_2/sep1"
}
layer {
  name: "conv4_2/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv4_2/sep1"
  top: "conv4_2/dw"
  convolution_param {
    num_output: 120
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv4_2/dw"
  top: "conv4_2/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_2/dw/scale"
  type: "Scale"
  bottom: "conv4_2/dw"
  top: "conv4_2/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_2/dw/ReLU"
  type: "ReLU"
  bottom: "conv4_2/dw"
  top: "conv4_2/dw"
}
layer {
  name: "conv4_2/sep2"
  type: "Convolution"
  bottom: "conv4_2/dw"
  top: "conv4_2/sep2"
  convolution_param {
    num_output: 40
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_2/sep2/bn"
  type: "BatchNorm"
  bottom: "conv4_2/sep2"
  top: "conv4_2/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_2/sep2/scale"
  type: "Scale"
  bottom: "conv4_2/sep2"
  top: "conv4_2/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_2/Eltwise1"
  type: "Eltwise"
  bottom: "conv4_1/sep2"
  bottom: "conv4_2/sep2"
  top: "conv4_2/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv4_2/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv4_2/Eltwise1"
  top: "conv4_2/Eltwise1"
}
layer {
  name: "conv4_3/sep1"
  type: "Convolution"
  bottom: "conv4_2/Eltwise1"
  top: "conv4_3/sep1"
  convolution_param {
    num_output: 120
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_3/sep1/bn"
  type: "BatchNorm"
  bottom: "conv4_3/sep1"
  top: "conv4_3/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_3/sep1/scale"
  type: "Scale"
  bottom: "conv4_3/sep1"
  top: "conv4_3/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_3/sep1/ReLU"
  type: "ReLU"
  bottom: "conv4_3/sep1"
  top: "conv4_3/sep1"
}
layer {
  name: "conv4_3/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv4_3/sep1"
  top: "conv4_3/dw"
  convolution_param {
    num_output: 120
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_3/dw/bn"
  type: "BatchNorm"
  bottom: "conv4_3/dw"
  top: "conv4_3/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_3/dw/scale"
  type: "Scale"
  bottom: "conv4_3/dw"
  top: "conv4_3/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_3/dw/ReLU"
  type: "ReLU"
  bottom: "conv4_3/dw"
  top: "conv4_3/dw"
}
layer {
  name: "conv4_3/sep2"
  type: "Convolution"
  bottom: "conv4_3/dw"
  top: "conv4_3/sep2"
  convolution_param {
    num_output: 40
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_3/sep2/bn"
  type: "BatchNorm"
  bottom: "conv4_3/sep2"
  top: "conv4_3/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv4_3/sep2/scale"
  type: "Scale"
  bottom: "conv4_3/sep2"
  top: "conv4_3/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_3/Eltwise1"
  type: "Eltwise"
  bottom: "conv4_2/Eltwise1"
  bottom: "conv4_3/sep2"
  top: "conv4_3/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv4_3/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv4_3/Eltwise1"
  top: "conv4_3/Eltwise1"
}
layer {
  name: "conv5_1/sep1"
  type: "Convolution"
  bottom: "conv4_3/Eltwise1"
  top: "conv5_1/sep1"
  convolution_param {
    num_output: 240
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/sep1/bn"
  type: "BatchNorm"
  bottom: "conv5_1/sep1"
  top: "conv5_1/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_1/sep1/scale"
  type: "Scale"
  bottom: "conv5_1/sep1"
  top: "conv5_1/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_1/sep1/ReLU"
  type: "ReLU"
  bottom: "conv5_1/sep1"
  top: "conv5_1/sep1"
}
layer {
  name: "conv5_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv5_1/sep1"
  top: "conv5_1/dw"
  convolution_param {
    num_output: 240
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_1/dw"
  top: "conv5_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_1/dw/scale"
  type: "Scale"
  bottom: "conv5_1/dw"
  top: "conv5_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv5_1/dw"
  top: "conv5_1/dw"
}
layer {
  name: "conv5_1/sep2"
  type: "Convolution"
  bottom: "conv5_1/dw"
  top: "conv5_1/sep2"
  convolution_param {
    num_output: 80
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/sep2/bn"
  type: "BatchNorm"
  bottom: "conv5_1/sep2"
  top: "conv5_1/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_1/sep2/scale"
  type: "Scale"
  bottom: "conv5_1/sep2"
  top: "conv5_1/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_1/sep2/ReLU"
  type: "ReLU"
  bottom: "conv5_1/sep2"
  top: "conv5_1/sep2"
}
layer {
  name: "conv5_2/sep1"
  type: "Convolution"
  bottom: "conv5_1/sep2"
  top: "conv5_2/sep1"
  convolution_param {
    num_output: 480
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_2/sep1/bn"
  type: "BatchNorm"
  bottom: "conv5_2/sep1"
  top: "conv5_2/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_2/sep1/scale"
  type: "Scale"
  bottom: "conv5_2/sep1"
  top: "conv5_2/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_2/sep1/ReLU"
  type: "ReLU"
  bottom: "conv5_2/sep1"
  top: "conv5_2/sep1"
}
layer {
  name: "conv5_2/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv5_2/sep1"
  top: "conv5_2/dw"
  convolution_param {
    num_output: 480
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_2/dw"
  top: "conv5_2/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_2/dw/scale"
  type: "Scale"
  bottom: "conv5_2/dw"
  top: "conv5_2/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_2/dw/ReLU"
  type: "ReLU"
  bottom: "conv5_2/dw"
  top: "conv5_2/dw"
}
layer {
  name: "conv5_2/sep2"
  type: "Convolution"
  bottom: "conv5_2/dw"
  top: "conv5_2/sep2"
  convolution_param {
    num_output: 80
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_2/sep2/bn"
  type: "BatchNorm"
  bottom: "conv5_2/sep2"
  top: "conv5_2/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_2/sep2/scale"
  type: "Scale"
  bottom: "conv5_2/sep2"
  top: "conv5_2/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_2/Eltwise1"
  type: "Eltwise"
  bottom: "conv5_1/sep2"
  bottom: "conv5_2/sep2"
  top: "conv5_2/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv5_2/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv5_2/Eltwise1"
  top: "conv5_2/Eltwise1"
}
layer {
  name: "conv5_3/sep1"
  type: "Convolution"
  bottom: "conv5_2/Eltwise1"
  top: "conv5_3/sep1"
  convolution_param {
    num_output: 480
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_3/sep1/bn"
  type: "BatchNorm"
  bottom: "conv5_3/sep1"
  top: "conv5_3/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_3/sep1/scale"
  type: "Scale"
  bottom: "conv5_3/sep1"
  top: "conv5_3/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_3/sep1/ReLU"
  type: "ReLU"
  bottom: "conv5_3/sep1"
  top: "conv5_3/sep1"
}
layer {
  name: "conv5_3/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv5_3/sep1"
  top: "conv5_3/dw"
  convolution_param {
    num_output: 480
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_3/dw/bn"
  type: "BatchNorm"
  bottom: "conv5_3/dw"
  top: "conv5_3/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_3/dw/scale"
  type: "Scale"
  bottom: "conv5_3/dw"
  top: "conv5_3/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_3/dw/ReLU"
  type: "ReLU"
  bottom: "conv5_3/dw"
  top: "conv5_3/dw"
}
layer {
  name: "conv5_3/sep2"
  type: "Convolution"
  bottom: "conv5_3/dw"
  top: "conv5_3/sep2"
  convolution_param {
    num_output: 80
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_3/sep2/bn"
  type: "BatchNorm"
  bottom: "conv5_3/sep2"
  top: "conv5_3/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv5_3/sep2/scale"
  type: "Scale"
  bottom: "conv5_3/sep2"
  top: "conv5_3/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_3/Eltwise1"
  type: "Eltwise"
  bottom: "conv5_2/Eltwise1"
  bottom: "conv5_3/sep2"
  top: "conv5_3/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv5_3/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv5_3/Eltwise1"
  top: "conv5_3/Eltwise1"
}
layer {
  name: "conv6_1/sep1"
  type: "Convolution"
  bottom: "conv5_3/Eltwise1"
  top: "conv6_1/sep1"
  convolution_param {
    num_output: 480
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6_1/sep1/bn"
  type: "BatchNorm"
  bottom: "conv6_1/sep1"
  top: "conv6_1/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv6_1/sep1/scale"
  type: "Scale"
  bottom: "conv6_1/sep1"
  top: "conv6_1/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv6_1/sep1/ReLU"
  type: "ReLU"
  bottom: "conv6_1/sep1"
  top: "conv6_1/sep1"
}
layer {
  name: "conv6_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv6_1/sep1"
  top: "conv6_1/dw"
  convolution_param {
    num_output: 480
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv6_1/dw"
  top: "conv6_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv6_1/dw/scale"
  type: "Scale"
  bottom: "conv6_1/dw"
  top: "conv6_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv6_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv6_1/dw"
  top: "conv6_1/dw"
}
layer {
  name: "conv6_1/sep2"
  type: "Convolution"
  bottom: "conv6_1/dw"
  top: "conv6_1/sep2"
  convolution_param {
    num_output: 96
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6_1/sep2/bn"
  type: "BatchNorm"
  bottom: "conv6_1/sep2"
  top: "conv6_1/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv6_1/sep2/scale"
  type: "Scale"
  bottom: "conv6_1/sep2"
  top: "conv6_1/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv6_1/sep2/ReLU"
  type: "ReLU"
  bottom: "conv6_1/sep2"
  top: "conv6_1/sep2"
}
layer {
  name: "conv6_2/sep1"
  type: "Convolution"
  bottom: "conv6_1/sep2"
  top: "conv6_2/sep1"
  convolution_param {
    num_output: 576
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6_2/sep1/bn"
  type: "BatchNorm"
  bottom: "conv6_2/sep1"
  top: "conv6_2/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv6_2/sep1/scale"
  type: "Scale"
  bottom: "conv6_2/sep1"
  top: "conv6_2/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv6_2/sep1/ReLU"
  type: "ReLU"
  bottom: "conv6_2/sep1"
  top: "conv6_2/sep1"
}
layer {
  name: "conv6_2/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv6_2/sep1"
  top: "conv6_2/dw"
  convolution_param {
    num_output: 576
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv6_2/dw"
  top: "conv6_2/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv6_2/dw/scale"
  type: "Scale"
  bottom: "conv6_2/dw"
  top: "conv6_2/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv6_2/dw/ReLU"
  type: "ReLU"
  bottom: "conv6_2/dw"
  top: "conv6_2/dw"
}
layer {
  name: "conv6_2/sep2"
  type: "Convolution"
  bottom: "conv6_2/dw"
  top: "conv6_2/sep2"
  convolution_param {
    num_output: 96
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6_2/sep2/bn"
  type: "BatchNorm"
  bottom: "conv6_2/sep2"
  top: "conv6_2/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv6_2/sep2/scale"
  type: "Scale"
  bottom: "conv6_2/sep2"
  top: "conv6_2/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv6_2/Eltwise1"
  type: "Eltwise"
  bottom: "conv6_1/sep2"
  bottom: "conv6_2/sep2"
  top: "conv6_2/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv6_2/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv6_2/Eltwise1"
  top: "conv6_2/Eltwise1"
}
layer {
  name: "conv7_1/sep1"
  type: "Convolution"
  bottom: "conv6_2/Eltwise1"
  top: "conv7_1/sep1"
  convolution_param {
    num_output: 576
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_1/sep1/bn"
  type: "BatchNorm"
  bottom: "conv7_1/sep1"
  top: "conv7_1/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_1/sep1/scale"
  type: "Scale"
  bottom: "conv7_1/sep1"
  top: "conv7_1/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_1/sep1/ReLU"
  type: "ReLU"
  bottom: "conv7_1/sep1"
  top: "conv7_1/sep1"
}
layer {
  name: "conv7_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv7_1/sep1"
  top: "conv7_1/dw"
  convolution_param {
    num_output: 576
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv7_1/dw"
  top: "conv7_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_1/dw/scale"
  type: "Scale"
  bottom: "conv7_1/dw"
  top: "conv7_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv7_1/dw"
  top: "conv7_1/dw"
}
layer {
  name: "conv7_1/sep2"
  type: "Convolution"
  bottom: "conv7_1/dw"
  top: "conv7_1/sep2"
  convolution_param {
    num_output: 192
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_1/sep2/bn"
  type: "BatchNorm"
  bottom: "conv7_1/sep2"
  top: "conv7_1/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_1/sep2/scale"
  type: "Scale"
  bottom: "conv7_1/sep2"
  top: "conv7_1/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_1/sep2/ReLU"
  type: "ReLU"
  bottom: "conv7_1/sep2"
  top: "conv7_1/sep2"
}
layer {
  name: "conv7_2/sep1"
  type: "Convolution"
  bottom: "conv7_1/sep2"
  top: "conv7_2/sep1"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_2/sep1/bn"
  type: "BatchNorm"
  bottom: "conv7_2/sep1"
  top: "conv7_2/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_2/sep1/scale"
  type: "Scale"
  bottom: "conv7_2/sep1"
  top: "conv7_2/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_2/sep1/ReLU"
  type: "ReLU"
  bottom: "conv7_2/sep1"
  top: "conv7_2/sep1"
}
layer {
  name: "conv7_2/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv7_2/sep1"
  top: "conv7_2/dw"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_2/dw/bn"
  type: "BatchNorm"
  bottom: "conv7_2/dw"
  top: "conv7_2/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_2/dw/scale"
  type: "Scale"
  bottom: "conv7_2/dw"
  top: "conv7_2/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_2/dw/ReLU"
  type: "ReLU"
  bottom: "conv7_2/dw"
  top: "conv7_2/dw"
}
layer {
  name: "conv7_2/sep2"
  type: "Convolution"
  bottom: "conv7_2/dw"
  top: "conv7_2/sep2"
  convolution_param {
    num_output: 192
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_2/sep2/bn"
  type: "BatchNorm"
  bottom: "conv7_2/sep2"
  top: "conv7_2/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_2/sep2/scale"
  type: "Scale"
  bottom: "conv7_2/sep2"
  top: "conv7_2/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_2/Eltwise1"
  type: "Eltwise"
  bottom: "conv7_1/sep2"
  bottom: "conv7_2/sep2"
  top: "conv7_2/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv7_2/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv7_2/Eltwise1"
  top: "conv7_2/Eltwise1"
}
layer {
  name: "conv7_3/sep1"
  type: "Convolution"
  bottom: "conv7_2/Eltwise1"
  top: "conv7_3/sep1"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_3/sep1/bn"
  type: "BatchNorm"
  bottom: "conv7_3/sep1"
  top: "conv7_3/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_3/sep1/scale"
  type: "Scale"
  bottom: "conv7_3/sep1"
  top: "conv7_3/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_3/sep1/ReLU"
  type: "ReLU"
  bottom: "conv7_3/sep1"
  top: "conv7_3/sep1"
}
layer {
  name: "conv7_3/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv7_3/sep1"
  top: "conv7_3/dw"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_3/dw/bn"
  type: "BatchNorm"
  bottom: "conv7_3/dw"
  top: "conv7_3/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_3/dw/scale"
  type: "Scale"
  bottom: "conv7_3/dw"
  top: "conv7_3/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_3/dw/ReLU"
  type: "ReLU"
  bottom: "conv7_3/dw"
  top: "conv7_3/dw"
}
layer {
  name: "conv7_3/sep2"
  type: "Convolution"
  bottom: "conv7_3/dw"
  top: "conv7_3/sep2"
  convolution_param {
    num_output: 192
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_3/sep2/bn"
  type: "BatchNorm"
  bottom: "conv7_3/sep2"
  top: "conv7_3/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_3/sep2/scale"
  type: "Scale"
  bottom: "conv7_3/sep2"
  top: "conv7_3/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_3/Eltwise1"
  type: "Eltwise"
  bottom: "conv7_2/Eltwise1"
  bottom: "conv7_3/sep2"
  top: "conv7_3/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv7_3/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv7_3/Eltwise1"
  top: "conv7_3/Eltwise1"
}
layer {
  name: "conv7_4/sep1"
  type: "Convolution"
  bottom: "conv7_3/Eltwise1"
  top: "conv7_4/sep1"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_4/sep1/bn"
  type: "BatchNorm"
  bottom: "conv7_4/sep1"
  top: "conv7_4/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_4/sep1/scale"
  type: "Scale"
  bottom: "conv7_4/sep1"
  top: "conv7_4/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_4/sep1/ReLU"
  type: "ReLU"
  bottom: "conv7_4/sep1"
  top: "conv7_4/sep1"
}
layer {
  name: "conv7_4/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv7_4/sep1"
  top: "conv7_4/dw"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 2
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_4/dw/bn"
  type: "BatchNorm"
  bottom: "conv7_4/dw"
  top: "conv7_4/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_4/dw/scale"
  type: "Scale"
  bottom: "conv7_4/dw"
  top: "conv7_4/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_4/dw/ReLU"
  type: "ReLU"
  bottom: "conv7_4/dw"
  top: "conv7_4/dw"
}
layer {
  name: "conv7_4/sep2"
  type: "Convolution"
  bottom: "conv7_4/dw"
  top: "conv7_4/sep2"
  convolution_param {
    num_output: 192
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7_4/sep2/bn"
  type: "BatchNorm"
  bottom: "conv7_4/sep2"
  top: "conv7_4/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv7_4/sep2/scale"
  type: "Scale"
  bottom: "conv7_4/sep2"
  top: "conv7_4/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv7_4/Eltwise1"
  type: "Eltwise"
  bottom: "conv7_3/Eltwise1"
  bottom: "conv7_4/sep2"
  top: "conv7_4/Eltwise1"
  eltwise_param {
    operation: SUM
  }
}
layer {
  name: "conv7_4/Eltwise/ReLU"
  type: "ReLU"
  bottom: "conv7_4/Eltwise1"
  top: "conv7_4/Eltwise1"
}
layer {
  name: "conv8_1/sep1"
  type: "Convolution"
  bottom: "conv7_4/Eltwise1"
  top: "conv8_1/sep1"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv8_1/sep1/bn"
  type: "BatchNorm"
  bottom: "conv8_1/sep1"
  top: "conv8_1/sep1"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv8_1/sep1/scale"
  type: "Scale"
  bottom: "conv8_1/sep1"
  top: "conv8_1/sep1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv8_1/sep1/ReLU"
  type: "ReLU"
  bottom: "conv8_1/sep1"
  top: "conv8_1/sep1"
}
layer {
  name: "conv8_1/dw"
  type: "ConvolutionDepthwise"
  bottom: "conv8_1/sep1"
  top: "conv8_1/dw"
  convolution_param {
    num_output: 1152
    bias_term: False
    pad: 1
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv8_1/dw/bn"
  type: "BatchNorm"
  bottom: "conv8_1/dw"
  top: "conv8_1/dw"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv8_1/dw/scale"
  type: "Scale"
  bottom: "conv8_1/dw"
  top: "conv8_1/dw"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv8_1/dw/ReLU"
  type: "ReLU"
  bottom: "conv8_1/dw"
  top: "conv8_1/dw"
}
layer {
  name: "conv8_1/sep2"
  type: "Convolution"
  bottom: "conv8_1/dw"
  top: "conv8_1/sep2"
  convolution_param {
    num_output: 320
    bias_term: False
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv8_1/sep2/bn"
  type: "BatchNorm"
  bottom: "conv8_1/sep2"
  top: "conv8_1/sep2"
  batch_norm_param {
    use_global_stats: True
  }
}
layer {
  name: "conv8_1/sep2/scale"
  type: "Scale"
  bottom: "conv8_1/sep2"
  top: "conv8_1/sep2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv8_1/sep2/ReLU"
  type: "ReLU"
  bottom: "conv8_1/sep2"
  top: "conv8_1/sep2"
}
layer {
  name: "Pooling1"
  type: "Pooling"
  bottom: "conv8_1/sep2"
  top: "Pooling1"
  pooling_param {
    pool: AVE
    global_pooling: True
  }
}
layer {
  name: "fc1"
  type: "InnerProduct"
  bottom: "Pooling1"
  top: "fc1"
  inner_product_param {
    num_output: 2
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}
layer {
  name: "prob"
  type: "Softmax"
  bottom: "fc1"
  top: "prob"
}