deploy_custom.prototxt

name: "MobileNetv2-YOLOv2Loss"
input: "data"
input_shape {
  dim: 1
  dim: 3
  dim: 416
  dim: 416
}
layer {
  name: "Conv"
  type: "Convolution"
  bottom: "data"
  top: "Conv"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 32
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "Conv/bn"
  type: "BatchNorm"
  bottom: "Conv"
  top: "Conv"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "Conv/scale"
  type: "Scale"
  bottom: "Conv"
  top: "Conv"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "Conv/relu"
  type: "ReLU6"
  bottom: "Conv"
  top: "Conv"
}
layer {
  name: "conv/depthwise"
  type: "Convolution"
  bottom: "Conv"
  top: "conv/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 32
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 32
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv/depthwise"
  top: "conv/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv/depthwise/scale"
  type: "Scale"
  bottom: "conv/depthwise"
  top: "conv/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv/depthwise/relu"
  type: "ReLU6"
  bottom: "conv/depthwise"
  top: "conv/depthwise"
}
layer {
  name: "conv/project"
  type: "Convolution"
  bottom: "conv/depthwise"
  top: "conv/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 16
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv/project/bn"
  type: "BatchNorm"
  bottom: "conv/project"
  top: "conv/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv/project/scale"
  type: "Scale"
  bottom: "conv/project"
  top: "conv/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_1/expand"
  type: "Convolution"
  bottom: "conv/project"
  top: "conv_1/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_1/expand/bn"
  type: "BatchNorm"
  bottom: "conv_1/expand"
  top: "conv_1/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_1/expand/scale"
  type: "Scale"
  bottom: "conv_1/expand"
  top: "conv_1/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_1/expand/relu"
  type: "ReLU6"
  bottom: "conv_1/expand"
  top: "conv_1/expand"
}
layer {
  name: "conv_1/depthwise"
  type: "Convolution"
  bottom: "conv_1/expand"
  top: "conv_1/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 96
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    group: 96
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_1/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_1/depthwise"
  top: "conv_1/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_1/depthwise/scale"
  type: "Scale"
  bottom: "conv_1/depthwise"
  top: "conv_1/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_1/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_1/depthwise"
  top: "conv_1/depthwise"
}
layer {
  name: "conv_1/project"
  type: "Convolution"
  bottom: "conv_1/depthwise"
  top: "conv_1/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 24
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_1/project/bn"
  type: "BatchNorm"
  bottom: "conv_1/project"
  top: "conv_1/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_1/project/scale"
  type: "Scale"
  bottom: "conv_1/project"
  top: "conv_1/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_2/expand"
  type: "Convolution"
  bottom: "conv_1/project"
  top: "conv_2/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 144
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_2/expand/bn"
  type: "BatchNorm"
  bottom: "conv_2/expand"
  top: "conv_2/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_2/expand/scale"
  type: "Scale"
  bottom: "conv_2/expand"
  top: "conv_2/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_2/expand/relu"
  type: "ReLU6"
  bottom: "conv_2/expand"
  top: "conv_2/expand"
}
layer {
  name: "conv_2/depthwise"
  type: "Convolution"
  bottom: "conv_2/expand"
  top: "conv_2/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 144
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 144
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_2/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_2/depthwise"
  top: "conv_2/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_2/depthwise/scale"
  type: "Scale"
  bottom: "conv_2/depthwise"
  top: "conv_2/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_2/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_2/depthwise"
  top: "conv_2/depthwise"
}
layer {
  name: "conv_2/project"
  type: "Convolution"
  bottom: "conv_2/depthwise"
  top: "conv_2/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 24
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_2/project/bn"
  type: "BatchNorm"
  bottom: "conv_2/project"
  top: "conv_2/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_2/project/scale"
  type: "Scale"
  bottom: "conv_2/project"
  top: "conv_2/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_2/sum"
  type: "Eltwise"
  bottom: "conv_1/project"
  bottom: "conv_2/project"
  top: "conv_2"
}
layer {
  name: "conv_3/expand"
  type: "Convolution"
  bottom: "conv_2"
  top: "conv_3/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 144
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_3/expand/bn"
  type: "BatchNorm"
  bottom: "conv_3/expand"
  top: "conv_3/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_3/expand/scale"
  type: "Scale"
  bottom: "conv_3/expand"
  top: "conv_3/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_3/expand/relu"
  type: "ReLU6"
  bottom: "conv_3/expand"
  top: "conv_3/expand"
}
layer {
  name: "conv_3/depthwise"
  type: "Convolution"
  bottom: "conv_3/expand"
  top: "conv_3/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 144
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    group: 144
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_3/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_3/depthwise"
  top: "conv_3/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_3/depthwise/scale"
  type: "Scale"
  bottom: "conv_3/depthwise"
  top: "conv_3/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_3/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_3/depthwise"
  top: "conv_3/depthwise"
}
layer {
  name: "conv_3/project"
  type: "Convolution"
  bottom: "conv_3/depthwise"
  top: "conv_3/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_3/project/bn"
  type: "BatchNorm"
  bottom: "conv_3/project"
  top: "conv_3/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_3/project/scale"
  type: "Scale"
  bottom: "conv_3/project"
  top: "conv_3/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_4/expand"
  type: "Convolution"
  bottom: "conv_3/project"
  top: "conv_4/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 192
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_4/expand/bn"
  type: "BatchNorm"
  bottom: "conv_4/expand"
  top: "conv_4/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_4/expand/scale"
  type: "Scale"
  bottom: "conv_4/expand"
  top: "conv_4/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_4/expand/relu"
  type: "ReLU6"
  bottom: "conv_4/expand"
  top: "conv_4/expand"
}
layer {
  name: "conv_4/depthwise"
  type: "Convolution"
  bottom: "conv_4/expand"
  top: "conv_4/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 192
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 192
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_4/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_4/depthwise"
  top: "conv_4/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_4/depthwise/scale"
  type: "Scale"
  bottom: "conv_4/depthwise"
  top: "conv_4/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_4/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_4/depthwise"
  top: "conv_4/depthwise"
}
layer {
  name: "conv_4/project"
  type: "Convolution"
  bottom: "conv_4/depthwise"
  top: "conv_4/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_4/project/bn"
  type: "BatchNorm"
  bottom: "conv_4/project"
  top: "conv_4/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_4/project/scale"
  type: "Scale"
  bottom: "conv_4/project"
  top: "conv_4/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_4/sum"
  type: "Eltwise"
  bottom: "conv_3/project"
  bottom: "conv_4/project"
  top: "conv_4"
}
layer {
  name: "conv_5/expand"
  type: "Convolution"
  bottom: "conv_4"
  top: "conv_5/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 192
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_5/expand/bn"
  type: "BatchNorm"
  bottom: "conv_5/expand"
  top: "conv_5/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_5/expand/scale"
  type: "Scale"
  bottom: "conv_5/expand"
  top: "conv_5/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_5/expand/relu"
  type: "ReLU6"
  bottom: "conv_5/expand"
  top: "conv_5/expand"
}
layer {
  name: "conv_5/depthwise"
  type: "Convolution"
  bottom: "conv_5/expand"
  top: "conv_5/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 192
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 192
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_5/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_5/depthwise"
  top: "conv_5/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_5/depthwise/scale"
  type: "Scale"
  bottom: "conv_5/depthwise"
  top: "conv_5/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_5/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_5/depthwise"
  top: "conv_5/depthwise"
}
layer {
  name: "conv_5/project"
  type: "Convolution"
  bottom: "conv_5/depthwise"
  top: "conv_5/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_5/project/bn"
  type: "BatchNorm"
  bottom: "conv_5/project"
  top: "conv_5/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_5/project/scale"
  type: "Scale"
  bottom: "conv_5/project"
  top: "conv_5/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_5/sum"
  type: "Eltwise"
  bottom: "conv_4"
  bottom: "conv_5/project"
  top: "conv_5"
}
layer {
  name: "conv_6/expand"
  type: "Convolution"
  bottom: "conv_5"
  top: "conv_6/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 192
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_6/expand/bn"
  type: "BatchNorm"
  bottom: "conv_6/expand"
  top: "conv_6/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_6/expand/scale"
  type: "Scale"
  bottom: "conv_6/expand"
  top: "conv_6/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_6/expand/relu"
  type: "ReLU6"
  bottom: "conv_6/expand"
  top: "conv_6/expand"
}
layer {
  name: "conv_6/depthwise"
  type: "Convolution"
  bottom: "conv_6/expand"
  top: "conv_6/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 192
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    group: 192
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_6/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_6/depthwise"
  top: "conv_6/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_6/depthwise/scale"
  type: "Scale"
  bottom: "conv_6/depthwise"
  top: "conv_6/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_6/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_6/depthwise"
  top: "conv_6/depthwise"
}
layer {
  name: "conv_6/project"
  type: "Convolution"
  bottom: "conv_6/depthwise"
  top: "conv_6/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_6/project/bn"
  type: "BatchNorm"
  bottom: "conv_6/project"
  top: "conv_6/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_6/project/scale"
  type: "Scale"
  bottom: "conv_6/project"
  top: "conv_6/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_7/expand"
  type: "Convolution"
  bottom: "conv_6/project"
  top: "conv_7/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_7/expand/bn"
  type: "BatchNorm"
  bottom: "conv_7/expand"
  top: "conv_7/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_7/expand/scale"
  type: "Scale"
  bottom: "conv_7/expand"
  top: "conv_7/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_7/expand/relu"
  type: "ReLU6"
  bottom: "conv_7/expand"
  top: "conv_7/expand"
}
layer {
  name: "conv_7/depthwise"
  type: "Convolution"
  bottom: "conv_7/expand"
  top: "conv_7/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_7/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_7/depthwise"
  top: "conv_7/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_7/depthwise/scale"
  type: "Scale"
  bottom: "conv_7/depthwise"
  top: "conv_7/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_7/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_7/depthwise"
  top: "conv_7/depthwise"
}
layer {
  name: "conv_7/project"
  type: "Convolution"
  bottom: "conv_7/depthwise"
  top: "conv_7/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_7/project/bn"
  type: "BatchNorm"
  bottom: "conv_7/project"
  top: "conv_7/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_7/project/scale"
  type: "Scale"
  bottom: "conv_7/project"
  top: "conv_7/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_7/sum"
  type: "Eltwise"
  bottom: "conv_6/project"
  bottom: "conv_7/project"
  top: "conv_7"
}
layer {
  name: "conv_8/expand"
  type: "Convolution"
  bottom: "conv_7"
  top: "conv_8/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_8/expand/bn"
  type: "BatchNorm"
  bottom: "conv_8/expand"
  top: "conv_8/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_8/expand/scale"
  type: "Scale"
  bottom: "conv_8/expand"
  top: "conv_8/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_8/expand/relu"
  type: "ReLU6"
  bottom: "conv_8/expand"
  top: "conv_8/expand"
}
layer {
  name: "conv_8/depthwise"
  type: "Convolution"
  bottom: "conv_8/expand"
  top: "conv_8/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_8/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_8/depthwise"
  top: "conv_8/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_8/depthwise/scale"
  type: "Scale"
  bottom: "conv_8/depthwise"
  top: "conv_8/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_8/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_8/depthwise"
  top: "conv_8/depthwise"
}
layer {
  name: "conv_8/project"
  type: "Convolution"
  bottom: "conv_8/depthwise"
  top: "conv_8/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_8/project/bn"
  type: "BatchNorm"
  bottom: "conv_8/project"
  top: "conv_8/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_8/project/scale"
  type: "Scale"
  bottom: "conv_8/project"
  top: "conv_8/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_8/sum"
  type: "Eltwise"
  bottom: "conv_7"
  bottom: "conv_8/project"
  top: "conv_8"
}
layer {
  name: "conv_9/expand"
  type: "Convolution"
  bottom: "conv_8"
  top: "conv_9/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_9/expand/bn"
  type: "BatchNorm"
  bottom: "conv_9/expand"
  top: "conv_9/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_9/expand/scale"
  type: "Scale"
  bottom: "conv_9/expand"
  top: "conv_9/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_9/expand/relu"
  type: "ReLU6"
  bottom: "conv_9/expand"
  top: "conv_9/expand"
}
layer {
  name: "conv_9/depthwise"
  type: "Convolution"
  bottom: "conv_9/expand"
  top: "conv_9/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_9/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_9/depthwise"
  top: "conv_9/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_9/depthwise/scale"
  type: "Scale"
  bottom: "conv_9/depthwise"
  top: "conv_9/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_9/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_9/depthwise"
  top: "conv_9/depthwise"
}
layer {
  name: "conv_9/project"
  type: "Convolution"
  bottom: "conv_9/depthwise"
  top: "conv_9/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_9/project/bn"
  type: "BatchNorm"
  bottom: "conv_9/project"
  top: "conv_9/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_9/project/scale"
  type: "Scale"
  bottom: "conv_9/project"
  top: "conv_9/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_9/sum"
  type: "Eltwise"
  bottom: "conv_8"
  bottom: "conv_9/project"
  top: "conv_9"
}
layer {
  name: "conv_10/expand"
  type: "Convolution"
  bottom: "conv_9"
  top: "conv_10/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_10/expand/bn"
  type: "BatchNorm"
  bottom: "conv_10/expand"
  top: "conv_10/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_10/expand/scale"
  type: "Scale"
  bottom: "conv_10/expand"
  top: "conv_10/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_10/expand/relu"
  type: "ReLU6"
  bottom: "conv_10/expand"
  top: "conv_10/expand"
}
layer {
  name: "conv_10/depthwise"
  type: "Convolution"
  bottom: "conv_10/expand"
  top: "conv_10/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_10/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_10/depthwise"
  top: "conv_10/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_10/depthwise/scale"
  type: "Scale"
  bottom: "conv_10/depthwise"
  top: "conv_10/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_10/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_10/depthwise"
  top: "conv_10/depthwise"
}
layer {
  name: "conv_10/project"
  type: "Convolution"
  bottom: "conv_10/depthwise"
  top: "conv_10/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_10/project/bn"
  type: "BatchNorm"
  bottom: "conv_10/project"
  top: "conv_10/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_10/project/scale"
  type: "Scale"
  bottom: "conv_10/project"
  top: "conv_10/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_11/expand"
  type: "Convolution"
  bottom: "conv_10/project"
  top: "conv_11/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 576
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_11/expand/bn"
  type: "BatchNorm"
  bottom: "conv_11/expand"
  top: "conv_11/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_11/expand/scale"
  type: "Scale"
  bottom: "conv_11/expand"
  top: "conv_11/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_11/expand/relu"
  type: "ReLU6"
  bottom: "conv_11/expand"
  top: "conv_11/expand"
}
layer {
  name: "conv_11/depthwise"
  type: "Convolution"
  bottom: "conv_11/expand"
  top: "conv_11/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 576
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 576
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_11/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_11/depthwise"
  top: "conv_11/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_11/depthwise/scale"
  type: "Scale"
  bottom: "conv_11/depthwise"
  top: "conv_11/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_11/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_11/depthwise"
  top: "conv_11/depthwise"
}
layer {
  name: "conv_11/project"
  type: "Convolution"
  bottom: "conv_11/depthwise"
  top: "conv_11/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_11/project/bn"
  type: "BatchNorm"
  bottom: "conv_11/project"
  top: "conv_11/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_11/project/scale"
  type: "Scale"
  bottom: "conv_11/project"
  top: "conv_11/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_11/sum"
  type: "Eltwise"
  bottom: "conv_10/project"
  bottom: "conv_11/project"
  top: "conv_11"
}
layer {
  name: "conv_12/expand"
  type: "Convolution"
  bottom: "conv_11"
  top: "conv_12/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 576
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_12/expand/bn"
  type: "BatchNorm"
  bottom: "conv_12/expand"
  top: "conv_12/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_12/expand/scale"
  type: "Scale"
  bottom: "conv_12/expand"
  top: "conv_12/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_12/expand/relu"
  type: "ReLU6"
  bottom: "conv_12/expand"
  top: "conv_12/expand"
}
layer {
  name: "conv_12/depthwise"
  type: "Convolution"
  bottom: "conv_12/expand"
  top: "conv_12/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 576
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 576
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_12/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_12/depthwise"
  top: "conv_12/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_12/depthwise/scale"
  type: "Scale"
  bottom: "conv_12/depthwise"
  top: "conv_12/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_12/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_12/depthwise"
  top: "conv_12/depthwise"
}
layer {
  name: "conv_12/project"
  type: "Convolution"
  bottom: "conv_12/depthwise"
  top: "conv_12/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_12/project/bn"
  type: "BatchNorm"
  bottom: "conv_12/project"
  top: "conv_12/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_12/project/scale"
  type: "Scale"
  bottom: "conv_12/project"
  top: "conv_12/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_12/sum"
  type: "Eltwise"
  bottom: "conv_11"
  bottom: "conv_12/project"
  top: "conv_12"
}
layer {
  name: "conv_13/expand"
  type: "Convolution"
  bottom: "conv_12"
  top: "conv_13/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 576
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_13/expand/bn"
  type: "BatchNorm"
  bottom: "conv_13/expand"
  top: "conv_13/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_13/expand/scale"
  type: "Scale"
  bottom: "conv_13/expand"
  top: "conv_13/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_13/expand/relu"
  type: "ReLU6"
  bottom: "conv_13/expand"
  top: "conv_13/expand"
}
layer {
  name: "conv_13/depthwise"
  type: "Convolution"
  bottom: "conv_13/expand"
  top: "conv_13/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 576
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 1
    group: 576
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_13/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_13/depthwise"
  top: "conv_13/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_13/depthwise/scale"
  type: "Scale"
  bottom: "conv_13/depthwise"
  top: "conv_13/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_13/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_13/depthwise"
  top: "conv_13/depthwise"
}
layer {
  name: "conv_13/project"
  type: "Convolution"
  bottom: "conv_13/depthwise"
  top: "conv_13/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 160
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_13/project/bn"
  type: "BatchNorm"
  bottom: "conv_13/project"
  top: "conv_13/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_13/project/scale"
  type: "Scale"
  bottom: "conv_13/project"
  top: "conv_13/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_14/expand"
  type: "Convolution"
  bottom: "conv_13/project"
  top: "conv_14/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 960
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_14/expand/bn"
  type: "BatchNorm"
  bottom: "conv_14/expand"
  top: "conv_14/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_14/expand/scale"
  type: "Scale"
  bottom: "conv_14/expand"
  top: "conv_14/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_14/expand/relu"
  type: "ReLU6"
  bottom: "conv_14/expand"
  top: "conv_14/expand"
}
layer {
  name: "conv_14/depthwise"
  type: "Convolution"
  bottom: "conv_14/expand"
  top: "conv_14/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 960
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 960
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_14/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_14/depthwise"
  top: "conv_14/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_14/depthwise/scale"
  type: "Scale"
  bottom: "conv_14/depthwise"
  top: "conv_14/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_14/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_14/depthwise"
  top: "conv_14/depthwise"
}
layer {
  name: "conv_14/project"
  type: "Convolution"
  bottom: "conv_14/depthwise"
  top: "conv_14/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 160
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_14/project/bn"
  type: "BatchNorm"
  bottom: "conv_14/project"
  top: "conv_14/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_14/project/scale"
  type: "Scale"
  bottom: "conv_14/project"
  top: "conv_14/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_14/sum"
  type: "Eltwise"
  bottom: "conv_13/project"
  bottom: "conv_14/project"
  top: "conv_14"
}
layer {
  name: "conv_15/expand"
  type: "Convolution"
  bottom: "conv_14"
  top: "conv_15/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 960
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_15/expand/bn"
  type: "BatchNorm"
  bottom: "conv_15/expand"
  top: "conv_15/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_15/expand/scale"
  type: "Scale"
  bottom: "conv_15/expand"
  top: "conv_15/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_15/expand/relu"
  type: "ReLU6"
  bottom: "conv_15/expand"
  top: "conv_15/expand"
}
layer {
  name: "conv_15/depthwise"
  type: "Convolution"
  bottom: "conv_15/expand"
  top: "conv_15/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 960
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 960
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_15/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_15/depthwise"
  top: "conv_15/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_15/depthwise/scale"
  type: "Scale"
  bottom: "conv_15/depthwise"
  top: "conv_15/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_15/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_15/depthwise"
  top: "conv_15/depthwise"
}
layer {
  name: "conv_15/project"
  type: "Convolution"
  bottom: "conv_15/depthwise"
  top: "conv_15/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 160
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_15/project/bn"
  type: "BatchNorm"
  bottom: "conv_15/project"
  top: "conv_15/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_15/project/scale"
  type: "Scale"
  bottom: "conv_15/project"
  top: "conv_15/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_15/sum"
  type: "Eltwise"
  bottom: "conv_14"
  bottom: "conv_15/project"
  top: "conv_15"
}
layer {
  name: "conv_16/expand"
  type: "Convolution"
  bottom: "conv_15"
  top: "conv_16/expand"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 960
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_16/expand/bn"
  type: "BatchNorm"
  bottom: "conv_16/expand"
  top: "conv_16/expand"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_16/expand/scale"
  type: "Scale"
  bottom: "conv_16/expand"
  top: "conv_16/expand"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_16/expand/relu"
  type: "ReLU6"
  bottom: "conv_16/expand"
  top: "conv_16/expand"
}
layer {
  name: "conv_16/depthwise"
  type: "Convolution"
  bottom: "conv_16/expand"
  top: "conv_16/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 960
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 960
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_16/depthwise/bn"
  type: "BatchNorm"
  bottom: "conv_16/depthwise"
  top: "conv_16/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_16/depthwise/scale"
  type: "Scale"
  bottom: "conv_16/depthwise"
  top: "conv_16/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv_16/depthwise/relu"
  type: "ReLU6"
  bottom: "conv_16/depthwise"
  top: "conv_16/depthwise"
}
layer {
  name: "conv_16/project"
  type: "Convolution"
  bottom: "conv_16/depthwise"
  top: "conv_16/project"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 320
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv_16/project/bn"
  type: "BatchNorm"
  bottom: "conv_16/project"
  top: "conv_16/project"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv_16/project/scale"
  type: "Scale"
  bottom: "conv_16/project"
  top: "conv_16/project"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "Conv_1"
  type: "Convolution"
  bottom: "conv_16/project"
  top: "Conv_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1280
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "Conv_1/bn"
  type: "BatchNorm"
  bottom: "Conv_1"
  top: "Conv_1"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "Conv_1/scale"
  type: "Scale"
  bottom: "Conv_1"
  top: "Conv_1"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "Conv_1/relu"
  type: "ReLU6"
  bottom: "Conv_1"
  top: "Conv_1"
}
layer {
  name: "layer_16"
  type: "Convolution"
  bottom: "Conv_1"
  top: "layer_16"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_16/bn"
  type: "BatchNorm"
  bottom: "layer_16"
  top: "layer_16"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_16/scale"
  type: "Scale"
  bottom: "layer_16"
  top: "layer_16"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_16/relu"
  type: "ReLU6"
  bottom: "layer_16"
  top: "layer_16"
}
layer {
  name: "layer_17/depthwise"
  type: "Convolution"
  bottom: "layer_16"
  top: "layer_17/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 1
    group: 512
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_17/depthwise/bn"
  type: "BatchNorm"
  bottom: "layer_17/depthwise"
  top: "layer_17/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_17/depthwise/scale"
  type: "Scale"
  bottom: "layer_17/depthwise"
  top: "layer_17/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_17/depthwise/relu"
  type: "ReLU6"
  bottom: "layer_17/depthwise"
  top: "layer_17/depthwise"
}
layer {
  name: "layer_17"
  type: "Convolution"
  bottom: "layer_17/depthwise"
  top: "layer_17"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_17/bn"
  type: "BatchNorm"
  bottom: "layer_17"
  top: "layer_17"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_17/scale"
  type: "Scale"
  bottom: "layer_17"
  top: "layer_17"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_17/relu"
  type: "ReLU6"
  bottom: "layer_17"
  top: "layer_17"
}
layer {
  name: "layer_18"
  type: "Convolution"
  bottom: "layer_17"
  top: "layer_18"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_18/bn"
  type: "BatchNorm"
  bottom: "layer_18"
  top: "layer_18"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_18/scale"
  type: "Scale"
  bottom: "layer_18"
  top: "layer_18"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_18/relu"
  type: "ReLU6"
  bottom: "layer_18"
  top: "layer_18"
}
layer {
  name: "layer_19/depthwise"
  type: "Convolution"
  bottom: "layer_18"
  top: "layer_19/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 1
    group: 512
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_19/depthwise/bn"
  type: "BatchNorm"
  bottom: "layer_19/depthwise"
  top: "layer_19/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_19/depthwise/scale"
  type: "Scale"
  bottom: "layer_19/depthwise"
  top: "layer_19/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_19/depthwise/relu"
  type: "ReLU6"
  bottom: "layer_19/depthwise"
  top: "layer_19/depthwise"
}
layer {
  name: "layer_19"
  type: "Convolution"
  bottom: "layer_19/depthwise"
  top: "layer_19"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_19/bn"
  type: "BatchNorm"
  bottom: "layer_19"
  top: "layer_19"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_19/scale"
  type: "Scale"
  bottom: "layer_19"
  top: "layer_19"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_19/relu"
  type: "ReLU6"
  bottom: "layer_19"
  top: "layer_19"
}
layer {
  name: "layer_20"
  type: "Convolution"
  bottom: "layer_19"
  top: "layer_20"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_20/bn"
  type: "BatchNorm"
  bottom: "layer_20"
  top: "layer_20"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_20/scale"
  type: "Scale"
  bottom: "layer_20"
  top: "layer_20"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_20/relu"
  type: "ReLU6"
  bottom: "layer_20"
  top: "layer_20"
}
layer {
  name: "layer_21/depthwise"
  type: "Convolution"
  bottom: "layer_20"
  top: "layer_21/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 1
    group: 512
    engine: CAFFE
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_21/depthwise/bn"
  type: "BatchNorm"
  bottom: "layer_21/depthwise"
  top: "layer_21/depthwise"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_21/depthwise/scale"
  type: "Scale"
  bottom: "layer_21/depthwise"
  top: "layer_21/depthwise"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_21/depthwise/relu"
  type: "ReLU6"
  bottom: "layer_21/depthwise"
  top: "layer_21/depthwise"
}
layer {
  name: "layer_21"
  type: "Convolution"
  bottom: "layer_21/depthwise"
  top: "layer_21"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "layer_21/bn"
  type: "BatchNorm"
  bottom: "layer_21"
  top: "layer_21"
  batch_norm_param {
    eps: 0.001
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "layer_21/scale"
  type: "Scale"
  bottom: "layer_21"
  top: "layer_21"
  param {
    lr_mult: 1.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "layer_21/relu"
  type: "ReLU6"
  bottom: "layer_21"
  top: "layer_21"
}

layer {
  name: "conv22_indoor"
  type: "Convolution"
  bottom: "layer_21"
  top: "conv22"
    param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 50
    kernel_size: 1
    pad: 0
    stride: 1
    weight_filler {
      type: "xavier"
    }
	bias_filler {
      value: 0
    }
  }
}
layer {
  name: "detection_out"
  type: "YoloDetectionOutput"
  bottom: "conv22"
  top: "detection_out"
  include {
	phase: TEST
  }
  yolo_detection_output_param {
    num_classes: 5
    coords: 4
    confidence_threshold: 0.5
	nms_threshold: 0.45
	side : 26
    biases: 1.08
    biases: 1.19
    biases: 3.42
    biases: 4.41
    biases: 6.63
    biases: 11.38
    biases: 9.42
    biases: 5.11
    biases: 16.62
    biases: 10.52
  }
}