MobileNetSSDV2_deploy.prototxt

name: "MobileNetV2-SSD"
input: "data"
input_shape {
  dim: 1
  dim: 3
  dim: 300
  dim: 300
}
layer {
  name: "conv0"
  type: "Convolution"
  bottom: "data"
  top: "conv0"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 32
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv0/bn"
  type: "BatchNorm"
  bottom: "conv0"
  top: "conv0/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv0/scale"
  type: "Scale"
  bottom: "conv0/bn"
  top: "conv0/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv0/bn"
  top: "conv0/bn"
}
layer {
  name: "conv1/expand"
  type: "Convolution"
  bottom: "conv0/bn"
  top: "conv1/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv1/expand/bn"
  type: "BatchNorm"
  bottom: "conv1/expand"
  top: "conv1/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv1/expand/scale"
  type: "Scale"
  bottom: "conv1/expand/bn"
  top: "conv1/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_1/expand"
  type: "ReLU"
  bottom: "conv1/expand/bn"
  top: "conv1/expand/bn"
}
layer {
  name: "conv1/dwise"
  type: "Convolution"
  bottom: "conv1/expand/bn"
  top: "conv1/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 32
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 32
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv1/dwise/bn"
  type: "BatchNorm"
  bottom: "conv1/dwise"
  top: "conv1/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv1/dwise/scale"
  type: "Scale"
  bottom: "conv1/dwise/bn"
  top: "conv1/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_1/dwise"
  type: "ReLU"
  bottom: "conv1/dwise/bn"
  top: "conv1/dwise/bn"
}
layer {
  name: "conv1/linear"
  type: "Convolution"
  bottom: "conv1/dwise/bn"
  top: "conv1/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 16
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv1/linear/bn"
  type: "BatchNorm"
  bottom: "conv1/linear"
  top: "conv1/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv1/linear/scale"
  type: "Scale"
  bottom: "conv1/linear/bn"
  top: "conv1/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv2/expand"
  type: "Convolution"
  bottom: "conv1/linear/bn"
  top: "conv2/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2/expand/bn"
  type: "BatchNorm"
  bottom: "conv2/expand"
  top: "conv2/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv2/expand/scale"
  type: "Scale"
  bottom: "conv2/expand/bn"
  top: "conv2/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2/expand"
  type: "ReLU"
  bottom: "conv2/expand/bn"
  top: "conv2/expand/bn"
}
layer {
  name: "conv2/dwise"
  type: "Convolution"
  bottom: "conv2/expand/bn"
  top: "conv2/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 96
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 96
    stride: 2
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv2/dwise/bn"
  type: "BatchNorm"
  bottom: "conv2/dwise"
  top: "conv2/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv2/dwise/scale"
  type: "Scale"
  bottom: "conv2/dwise/bn"
  top: "conv2/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2/dwise"
  type: "ReLU"
  bottom: "conv2/dwise/bn"
  top: "conv2/dwise/bn"
}
layer {
  name: "conv2/linear"
  type: "Convolution"
  bottom: "conv2/dwise/bn"
  top: "conv2/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 24
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2/linear/bn"
  type: "BatchNorm"
  bottom: "conv2/linear"
  top: "conv2/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv2/linear/scale"
  type: "Scale"
  bottom: "conv2/linear/bn"
  top: "conv2/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3/expand"
  type: "Convolution"
  bottom: "conv2/linear/bn"
  top: "conv3/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 144
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3/expand/bn"
  type: "BatchNorm"
  bottom: "conv3/expand"
  top: "conv3/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv3/expand/scale"
  type: "Scale"
  bottom: "conv3/expand/bn"
  top: "conv3/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3/expand"
  type: "ReLU"
  bottom: "conv3/expand/bn"
  top: "conv3/expand/bn"
}
layer {
  name: "conv3/dwise"
  type: "Convolution"
  bottom: "conv3/expand/bn"
  top: "conv3/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 144
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 144
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv3/dwise/bn"
  type: "BatchNorm"
  bottom: "conv3/dwise"
  top: "conv3/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv3/dwise/scale"
  type: "Scale"
  bottom: "conv3/dwise/bn"
  top: "conv3/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3/dwise"
  type: "ReLU"
  bottom: "conv3/dwise/bn"
  top: "conv3/dwise/bn"
}
layer {
  name: "conv3/linear"
  type: "Convolution"
  bottom: "conv3/dwise/bn"
  top: "conv3/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 24
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3/linear/bn"
  type: "BatchNorm"
  bottom: "conv3/linear"
  top: "conv3/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv3/linear/scale"
  type: "Scale"
  bottom: "conv3/linear/bn"
  top: "conv3/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_3"
  type: "Eltwise"
  bottom: "conv2/linear/bn"
  bottom: "conv3/linear/bn"
  top: "block_3"
}
layer {
  name: "conv4/expand"
  type: "Convolution"
  bottom: "block_3"
  top: "conv4/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 144
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4/expand/bn"
  type: "BatchNorm"
  bottom: "conv4/expand"
  top: "conv4/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv4/expand/scale"
  type: "Scale"
  bottom: "conv4/expand/bn"
  top: "conv4/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4/expand"
  type: "ReLU"
  bottom: "conv4/expand/bn"
  top: "conv4/expand/bn"
}
layer {
  name: "conv4/dwise"
  type: "Convolution"
  bottom: "conv4/expand/bn"
  top: "conv4/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 144
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 144
    stride: 2
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv4/dwise/bn"
  type: "BatchNorm"
  bottom: "conv4/dwise"
  top: "conv4/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv4/dwise/scale"
  type: "Scale"
  bottom: "conv4/dwise/bn"
  top: "conv4/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4/dwise"
  type: "ReLU"
  bottom: "conv4/dwise/bn"
  top: "conv4/dwise/bn"
}
layer {
  name: "conv4/linear"
  type: "Convolution"
  bottom: "conv4/dwise/bn"
  top: "conv4/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4/linear/bn"
  type: "BatchNorm"
  bottom: "conv4/linear"
  top: "conv4/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv4/linear/scale"
  type: "Scale"
  bottom: "conv4/linear/bn"
  top: "conv4/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5/expand"
  type: "Convolution"
  bottom: "conv4/linear/bn"
  top: "conv5/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 192
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5/expand/bn"
  type: "BatchNorm"
  bottom: "conv5/expand"
  top: "conv5/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv5/expand/scale"
  type: "Scale"
  bottom: "conv5/expand/bn"
  top: "conv5/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5/expand"
  type: "ReLU"
  bottom: "conv5/expand/bn"
  top: "conv5/expand/bn"
}
layer {
  name: "conv5/dwise"
  type: "Convolution"
  bottom: "conv5/expand/bn"
  top: "conv5/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 192
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 192
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv5/dwise/bn"
  type: "BatchNorm"
  bottom: "conv5/dwise"
  top: "conv5/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv5/dwise/scale"
  type: "Scale"
  bottom: "conv5/dwise/bn"
  top: "conv5/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5/dwise"
  type: "ReLU"
  bottom: "conv5/dwise/bn"
  top: "conv5/dwise/bn"
}
layer {
  name: "conv5/linear"
  type: "Convolution"
  bottom: "conv5/dwise/bn"
  top: "conv5/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5/linear/bn"
  type: "BatchNorm"
  bottom: "conv5/linear"
  top: "conv5/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv5/linear/scale"
  type: "Scale"
  bottom: "conv5/linear/bn"
  top: "conv5/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_5"
  type: "Eltwise"
  bottom: "conv4/linear/bn"
  bottom: "conv5/linear/bn"
  top: "block_5"
}
layer {
  name: "conv6/expand"
  type: "Convolution"
  bottom: "block_5"
  top: "conv6/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 192
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6/expand/bn"
  type: "BatchNorm"
  bottom: "conv6/expand"
  top: "conv6/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv6/expand/scale"
  type: "Scale"
  bottom: "conv6/expand/bn"
  top: "conv6/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu6/expand"
  type: "ReLU"
  bottom: "conv6/expand/bn"
  top: "conv6/expand/bn"
}
layer {
  name: "conv6/dwise"
  type: "Convolution"
  bottom: "conv6/expand/bn"
  top: "conv6/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 192
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 192
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv6/dwise/bn"
  type: "BatchNorm"
  bottom: "conv6/dwise"
  top: "conv6/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv6/dwise/scale"
  type: "Scale"
  bottom: "conv6/dwise/bn"
  top: "conv6/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu6/dwise"
  type: "ReLU"
  bottom: "conv6/dwise/bn"
  top: "conv6/dwise/bn"
}
layer {
  name: "conv6/linear"
  type: "Convolution"
  bottom: "conv6/dwise/bn"
  top: "conv6/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 32
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv6/linear/bn"
  type: "BatchNorm"
  bottom: "conv6/linear"
  top: "conv6/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv6/linear/scale"
  type: "Scale"
  bottom: "conv6/linear/bn"
  top: "conv6/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_6"
  type: "Eltwise"
  bottom: "block_5"
  bottom: "conv6/linear/bn"
  top: "block_6"
}
layer {
  name: "conv7/expand"
  type: "Convolution"
  bottom: "block_6"
  top: "conv7/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 192
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7/expand/bn"
  type: "BatchNorm"
  bottom: "conv7/expand"
  top: "conv7/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv7/expand/scale"
  type: "Scale"
  bottom: "conv7/expand/bn"
  top: "conv7/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu7/expand"
  type: "ReLU"
  bottom: "conv7/expand/bn"
  top: "conv7/expand/bn"
}
layer {
  name: "conv7/dwise"
  type: "Convolution"
  bottom: "conv7/expand/bn"
  top: "conv7/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 192
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 192
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv7/dwise/bn"
  type: "BatchNorm"
  bottom: "conv7/dwise"
  top: "conv7/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv7/dwise/scale"
  type: "Scale"
  bottom: "conv7/dwise/bn"
  top: "conv7/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu7/dwise"
  type: "ReLU"
  bottom: "conv7/dwise/bn"
  top: "conv7/dwise/bn"
}
layer {
  name: "conv7/linear"
  type: "Convolution"
  bottom: "conv7/dwise/bn"
  top: "conv7/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv7/linear/bn"
  type: "BatchNorm"
  bottom: "conv7/linear"
  top: "conv7/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv7/linear/scale"
  type: "Scale"
  bottom: "conv7/linear/bn"
  top: "conv7/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv8/expand"
  type: "Convolution"
  bottom: "conv7/linear/bn"
  top: "conv8/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv8/expand/bn"
  type: "BatchNorm"
  bottom: "conv8/expand"
  top: "conv8/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv8/expand/scale"
  type: "Scale"
  bottom: "conv8/expand/bn"
  top: "conv8/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu8/expand"
  type: "ReLU"
  bottom: "conv8/expand/bn"
  top: "conv8/expand/bn"
}
layer {
  name: "conv8/dwise"
  type: "Convolution"
  bottom: "conv8/expand/bn"
  top: "conv8/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv8/dwise/bn"
  type: "BatchNorm"
  bottom: "conv8/dwise"
  top: "conv8/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv8/dwise/scale"
  type: "Scale"
  bottom: "conv8/dwise/bn"
  top: "conv8/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu8/dwise"
  type: "ReLU"
  bottom: "conv8/dwise/bn"
  top: "conv8/dwise/bn"
}
layer {
  name: "conv8/linear"
  type: "Convolution"
  bottom: "conv8/dwise/bn"
  top: "conv8/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv8/linear/bn"
  type: "BatchNorm"
  bottom: "conv8/linear"
  top: "conv8/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv8/linear/scale"
  type: "Scale"
  bottom: "conv8/linear/bn"
  top: "conv8/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_8"
  type: "Eltwise"
  bottom: "conv7/linear/bn"
  bottom: "conv8/linear/bn"
  top: "block_8"
}
layer {
  name: "conv9/expand"
  type: "Convolution"
  bottom: "block_8"
  top: "conv9/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv9/expand/bn"
  type: "BatchNorm"
  bottom: "conv9/expand"
  top: "conv9/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv9/expand/scale"
  type: "Scale"
  bottom: "conv9/expand/bn"
  top: "conv9/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu9/expand"
  type: "ReLU"
  bottom: "conv9/expand/bn"
  top: "conv9/expand/bn"
}
layer {
  name: "conv9/dwise"
  type: "Convolution"
  bottom: "conv9/expand/bn"
  top: "conv9/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv9/dwise/bn"
  type: "BatchNorm"
  bottom: "conv9/dwise"
  top: "conv9/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv9/dwise/scale"
  type: "Scale"
  bottom: "conv9/dwise/bn"
  top: "conv9/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu9/dwise"
  type: "ReLU"
  bottom: "conv9/dwise/bn"
  top: "conv9/dwise/bn"
}
layer {
  name: "conv9/linear"
  type: "Convolution"
  bottom: "conv9/dwise/bn"
  top: "conv9/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv9/linear/bn"
  type: "BatchNorm"
  bottom: "conv9/linear"
  top: "conv9/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv9/linear/scale"
  type: "Scale"
  bottom: "conv9/linear/bn"
  top: "conv9/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_9"
  type: "Eltwise"
  bottom: "block_8"
  bottom: "conv9/linear/bn"
  top: "block_9"
}
layer {
  name: "conv10/expand"
  type: "Convolution"
  bottom: "block_9"
  top: "conv10/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv10/expand/bn"
  type: "BatchNorm"
  bottom: "conv10/expand"
  top: "conv10/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv10/expand/scale"
  type: "Scale"
  bottom: "conv10/expand/bn"
  top: "conv10/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu10/expand"
  type: "ReLU"
  bottom: "conv10/expand/bn"
  top: "conv10/expand/bn"
}
layer {
  name: "conv10/dwise"
  type: "Convolution"
  bottom: "conv10/expand/bn"
  top: "conv10/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv10/dwise/bn"
  type: "BatchNorm"
  bottom: "conv10/dwise"
  top: "conv10/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv10/dwise/scale"
  type: "Scale"
  bottom: "conv10/dwise/bn"
  top: "conv10/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu10/dwise"
  type: "ReLU"
  bottom: "conv10/dwise/bn"
  top: "conv10/dwise/bn"
}
layer {
  name: "conv10/linear"
  type: "Convolution"
  bottom: "conv10/dwise/bn"
  top: "conv10/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv10/linear/bn"
  type: "BatchNorm"
  bottom: "conv10/linear"
  top: "conv10/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv10/linear/scale"
  type: "Scale"
  bottom: "conv10/linear/bn"
  top: "conv10/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_10"
  type: "Eltwise"
  bottom: "block_9"
  bottom: "conv10/linear/bn"
  top: "block_10"
}
layer {
  name: "conv11/expand"
  type: "Convolution"
  bottom: "block_10"
  top: "conv11/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv11/expand/bn"
  type: "BatchNorm"
  bottom: "conv11/expand"
  top: "conv11/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv11/expand/scale"
  type: "Scale"
  bottom: "conv11/expand/bn"
  top: "conv11/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu11/expand"
  type: "ReLU"
  bottom: "conv11/expand/bn"
  top: "conv11/expand/bn"
}
layer {
  name: "conv11/dwise"
  type: "Convolution"
  bottom: "conv11/expand/bn"
  top: "conv11/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 384
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 384
    stride: 2
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv11/dwise/bn"
  type: "BatchNorm"
  bottom: "conv11/dwise"
  top: "conv11/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv11/dwise/scale"
  type: "Scale"
  bottom: "conv11/dwise/bn"
  top: "conv11/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu11/dwise"
  type: "ReLU"
  bottom: "conv11/dwise/bn"
  top: "conv11/dwise/bn"
}
layer {
  name: "conv11/linear"
  type: "Convolution"
  bottom: "conv11/dwise/bn"
  top: "conv11/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv11/linear/bn"
  type: "BatchNorm"
  bottom: "conv11/linear"
  top: "conv11/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv11/linear/scale"
  type: "Scale"
  bottom: "conv11/linear/bn"
  top: "conv11/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv12/expand"
  type: "Convolution"
  bottom: "conv11/linear/bn"
  top: "conv12/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 576
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv12/expand/bn"
  type: "BatchNorm"
  bottom: "conv12/expand"
  top: "conv12/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv12/expand/scale"
  type: "Scale"
  bottom: "conv12/expand/bn"
  top: "conv12/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu12/expand"
  type: "ReLU"
  bottom: "conv12/expand/bn"
  top: "conv12/expand/bn"
}
layer {
  name: "conv12/dwise"
  type: "Convolution"
  bottom: "conv12/expand/bn"
  top: "conv12/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 576
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 576
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv12/dwise/bn"
  type: "BatchNorm"
  bottom: "conv12/dwise"
  top: "conv12/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv12/dwise/scale"
  type: "Scale"
  bottom: "conv12/dwise/bn"
  top: "conv12/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu12/dwise"
  type: "ReLU"
  bottom: "conv12/dwise/bn"
  top: "conv12/dwise/bn"
}
layer {
  name: "conv12/linear"
  type: "Convolution"
  bottom: "conv12/dwise/bn"
  top: "conv12/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv12/linear/bn"
  type: "BatchNorm"
  bottom: "conv12/linear"
  top: "conv12/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv12/linear/scale"
  type: "Scale"
  bottom: "conv12/linear/bn"
  top: "conv12/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_12"
  type: "Eltwise"
  bottom: "conv11/linear/bn"
  bottom: "conv12/linear/bn"
  top: "block_12"
}
layer {
  name: "conv13/expand"
  type: "Convolution"
  bottom: "block_12"
  top: "conv13/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 576
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv13/expand/bn"
  type: "BatchNorm"
  bottom: "conv13/expand"
  top: "conv13/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv13/expand/scale"
  type: "Scale"
  bottom: "conv13/expand/bn"
  top: "conv13/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu13/expand"
  type: "ReLU"
  bottom: "conv13/expand/bn"
  top: "conv13/expand/bn"
}
layer {
  name: "conv13/dwise"
  type: "Convolution"
  bottom: "conv13/expand/bn"
  top: "conv13/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 576
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 576
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv13/dwise/bn"
  type: "BatchNorm"
  bottom: "conv13/dwise"
  top: "conv13/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv13/dwise/scale"
  type: "Scale"
  bottom: "conv13/dwise/bn"
  top: "conv13/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu13/dwise"
  type: "ReLU"
  bottom: "conv13/dwise/bn"
  top: "conv13/dwise/bn"
}
layer {
  name: "conv13/linear"
  type: "Convolution"
  bottom: "conv13/dwise/bn"
  top: "conv13/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 96
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv13/linear/bn"
  type: "BatchNorm"
  bottom: "conv13/linear"
  top: "conv13/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv13/linear/scale"
  type: "Scale"
  bottom: "conv13/linear/bn"
  top: "conv13/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_13"
  type: "Eltwise"
  bottom: "block_12"
  bottom: "conv13/linear/bn"
  top: "block_13"
}
layer {
  name: "conv14/expand"
  type: "Convolution"
  bottom: "block_13"
  top: "conv14/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 576
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv14/expand/bn"
  type: "BatchNorm"
  bottom: "conv14/expand"
  top: "conv14/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv14/expand/scale"
  type: "Scale"
  bottom: "conv14/expand/bn"
  top: "conv14/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu14/expand"
  type: "ReLU"
  bottom: "conv14/expand/bn"
  top: "conv14/expand/bn"
}
layer {
  name: "conv14/dwise"
  type: "Convolution"
  bottom: "conv14/expand/bn"
  top: "conv14/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 576
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 576
    stride: 2
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv14/dwise/bn"
  type: "BatchNorm"
  bottom: "conv14/dwise"
  top: "conv14/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv14/dwise/scale"
  type: "Scale"
  bottom: "conv14/dwise/bn"
  top: "conv14/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu14/dwise"
  type: "ReLU"
  bottom: "conv14/dwise/bn"
  top: "conv14/dwise/bn"
}
layer {
  name: "conv14/linear"
  type: "Convolution"
  bottom: "conv14/dwise/bn"
  top: "conv14/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 160
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv14/linear/bn"
  type: "BatchNorm"
  bottom: "conv14/linear"
  top: "conv14/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv14/linear/scale"
  type: "Scale"
  bottom: "conv14/linear/bn"
  top: "conv14/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv15/expand"
  type: "Convolution"
  bottom: "conv14/linear/bn"
  top: "conv15/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 960
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv15/expand/bn"
  type: "BatchNorm"
  bottom: "conv15/expand"
  top: "conv15/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv15/expand/scale"
  type: "Scale"
  bottom: "conv15/expand/bn"
  top: "conv15/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu15/expand"
  type: "ReLU"
  bottom: "conv15/expand/bn"
  top: "conv15/expand/bn"
}
layer {
  name: "conv15/dwise"
  type: "Convolution"
  bottom: "conv15/expand/bn"
  top: "conv15/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 960
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 960
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv15/dwise/bn"
  type: "BatchNorm"
  bottom: "conv15/dwise"
  top: "conv15/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv15/dwise/scale"
  type: "Scale"
  bottom: "conv15/dwise/bn"
  top: "conv15/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu15/dwise"
  type: "ReLU"
  bottom: "conv15/dwise/bn"
  top: "conv15/dwise/bn"
}
layer {
  name: "conv15/linear"
  type: "Convolution"
  bottom: "conv15/dwise/bn"
  top: "conv15/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 160
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv15/linear/bn"
  type: "BatchNorm"
  bottom: "conv15/linear"
  top: "conv15/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv15/linear/scale"
  type: "Scale"
  bottom: "conv15/linear/bn"
  top: "conv15/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_15"
  type: "Eltwise"
  bottom: "conv14/linear/bn"
  bottom: "conv15/linear/bn"
  top: "block_15"
}
layer {
  name: "conv16/expand"
  type: "Convolution"
  bottom: "block_15"
  top: "conv16/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 960
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv16/expand/bn"
  type: "BatchNorm"
  bottom: "conv16/expand"
  top: "conv16/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv16/expand/scale"
  type: "Scale"
  bottom: "conv16/expand/bn"
  top: "conv16/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu16/expand"
  type: "ReLU"
  bottom: "conv16/expand/bn"
  top: "conv16/expand/bn"
}
layer {
  name: "conv16/dwise"
  type: "Convolution"
  bottom: "conv16/expand/bn"
  top: "conv16/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 960
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 960
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv16/dwise/bn"
  type: "BatchNorm"
  bottom: "conv16/dwise"
  top: "conv16/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv16/dwise/scale"
  type: "Scale"
  bottom: "conv16/dwise/bn"
  top: "conv16/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu16/dwise"
  type: "ReLU"
  bottom: "conv16/dwise/bn"
  top: "conv16/dwise/bn"
}
layer {
  name: "conv16/linear"
  type: "Convolution"
  bottom: "conv16/dwise/bn"
  top: "conv16/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 160
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv16/linear/bn"
  type: "BatchNorm"
  bottom: "conv16/linear"
  top: "conv16/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv16/linear/scale"
  type: "Scale"
  bottom: "conv16/linear/bn"
  top: "conv16/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "block_16"
  type: "Eltwise"
  bottom: "block_15"
  bottom: "conv16/linear/bn"
  top: "block_16"
}
layer {
  name: "conv17/expand"
  type: "Convolution"
  bottom: "block_16"
  top: "conv17/expand"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 960
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv17/expand/bn"
  type: "BatchNorm"
  bottom: "conv17/expand"
  top: "conv17/expand/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv17/expand/scale"
  type: "Scale"
  bottom: "conv17/expand/bn"
  top: "conv17/expand/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu17/expand"
  type: "ReLU"
  bottom: "conv17/expand/bn"
  top: "conv17/expand/bn"
}
layer {
  name: "conv17/dwise"
  type: "Convolution"
  bottom: "conv17/expand/bn"
  top: "conv17/dwise"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 960
    bias_term: false
    pad: 1
    kernel_size: 3
    group: 960
    weight_filler {
      type: "msra"
    }
    engine: CAFFE
  }
}
layer {
  name: "conv17/dwise/bn"
  type: "BatchNorm"
  bottom: "conv17/dwise"
  top: "conv17/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv17/dwise/scale"
  type: "Scale"
  bottom: "conv17/dwise/bn"
  top: "conv17/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu17/dwise"
  type: "ReLU"
  bottom: "conv17/dwise/bn"
  top: "conv17/dwise/bn"
}
layer {
  name: "conv17/linear"
  type: "Convolution"
  bottom: "conv17/dwise/bn"
  top: "conv17/linear"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 320
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv17/linear/bn"
  type: "BatchNorm"
  bottom: "conv17/linear"
  top: "conv17/linear/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv17/linear/scale"
  type: "Scale"
  bottom: "conv17/linear/bn"
  top: "conv17/linear/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv18"
  type: "Convolution"
  bottom: "conv17/linear/bn"
  top: "conv18"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 1280
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv18/bn"
  type: "BatchNorm"
  bottom: "conv18"
  top: "conv18/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv18/scale"
  type: "Scale"
  bottom: "conv18/bn"
  top: "conv18/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu18"
  type: "ReLU"
  bottom: "conv18/bn"
  top: "conv18/bn"
}
layer {
  name: "conv19_1"
  type: "Convolution"
  bottom: "conv18/bn"
  top: "conv19_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv19_1/bn"
  type: "BatchNorm"
  bottom: "conv19_1"
  top: "conv19_1/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv19_1/scale"
  type: "Scale"
  bottom: "conv19_1/bn"
  top: "conv19_1/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv19_1/relu"
  type: "ReLU"
  bottom: "conv19_1/bn"
  top: "conv19_1/bn"
}
layer {
  name: "conv19_2/dwise"
  type: "Convolution"
  bottom: "conv19_1/bn"
  top: "conv19_2/dwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 256
    stride: 2
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv19_2/dwise/bn"
  type: "BatchNorm"
  bottom: "conv19_2/dwise"
  top: "conv19_2/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv19_2/dwise/scale"
  type: "Scale"
  bottom: "conv19_2/dwise/bn"
  top: "conv19_2/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv19_2/dwise/relu"
  type: "ReLU"
  bottom: "conv19_2/dwise/bn"
  top: "conv19_2/dwise/bn"
}
layer {
  name: "conv19_3"
  type: "Convolution"
  bottom: "conv19_2/dwise/bn"
  top: "conv19_3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 512
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv19_3/bn"
  type: "BatchNorm"
  bottom: "conv19_3"
  top: "conv19_3/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv19_3/scale"
  type: "Scale"
  bottom: "conv19_3/bn"
  top: "conv19_3/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv19_3/relu"
  type: "ReLU"
  bottom: "conv19_3/bn"
  top: "conv19_3/bn"
}
layer {
  name: "conv20_1"
  type: "Convolution"
  bottom: "conv19_3/bn"
  top: "conv20_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 128
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv20_1/bn"
  type: "BatchNorm"
  bottom: "conv20_1"
  top: "conv20_1/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv20_1/scale"
  type: "Scale"
  bottom: "conv20_1/bn"
  top: "conv20_1/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv20_1/relu"
  type: "ReLU"
  bottom: "conv20_1/bn"
  top: "conv20_1/bn"
}
layer {
  name: "conv20_2/dwise"
  type: "Convolution"
  bottom: "conv20_1/bn"
  top: "conv20_2/dwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    group: 128
    stride: 2
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv20_2/dwise/bn"
  type: "BatchNorm"
  bottom: "conv20_2/dwise"
  top: "conv20_2/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv20_2/dwise/scale"
  type: "Scale"
  bottom: "conv20_2/dwise/bn"
  top: "conv20_2/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv20_2/dwise/relu"
  type: "ReLU"
  bottom: "conv20_2/dwise/bn"
  top: "conv20_2/dwise/bn"
}
layer {
  name: "conv20_3"
  type: "Convolution"
  bottom: "conv20_2/dwise/bn"
  top: "conv20_3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv20_3/bn"
  type: "BatchNorm"
  bottom: "conv20_3"
  top: "conv20_3/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv20_3/scale"
  type: "Scale"
  bottom: "conv20_3/bn"
  top: "conv20_3/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv20_3/relu"
  type: "ReLU"
  bottom: "conv20_3/bn"
  top: "conv20_3/bn"
}
layer {
  name: "conv21_1"
  type: "Convolution"
  bottom: "conv20_3/bn"
  top: "conv21_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 128
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv21_1/bn"
  type: "BatchNorm"
  bottom: "conv21_1"
  top: "conv21_1/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv21_1/scale"
  type: "Scale"
  bottom: "conv21_1/bn"
  top: "conv21_1/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv21_1/relu"
  type: "ReLU"
  bottom: "conv21_1/bn"
  top: "conv21_1/bn"
}
layer {
  name: "conv21_2/dwise"
  type: "Convolution"
  bottom: "conv21_1/bn"
  top: "conv21_2/dwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    group: 128
    stride: 2
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv21_2/dwise/bn"
  type: "BatchNorm"
  bottom: "conv21_2/dwise"
  top: "conv21_2/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv21_2/dwise/scale"
  type: "Scale"
  bottom: "conv21_2/dwise/bn"
  top: "conv21_2/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv21_2/dwise/relu"
  type: "ReLU"
  bottom: "conv21_2/dwise/bn"
  top: "conv21_2/dwise/bn"
}
layer {
  name: "conv21_3"
  type: "Convolution"
  bottom: "conv21_2/dwise/bn"
  top: "conv21_3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 256
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv21_3/bn"
  type: "BatchNorm"
  bottom: "conv21_3"
  top: "conv21_3/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv21_3/scale"
  type: "Scale"
  bottom: "conv21_3/bn"
  top: "conv21_3/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv21_3/relu"
  type: "ReLU"
  bottom: "conv21_3/bn"
  top: "conv21_3/bn"
}
layer {
  name: "conv22_1"
  type: "Convolution"
  bottom: "conv21_3/bn"
  top: "conv22_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 64
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv22_1/bn"
  type: "BatchNorm"
  bottom: "conv22_1"
  top: "conv22_1/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv22_1/scale"
  type: "Scale"
  bottom: "conv22_1/bn"
  top: "conv22_1/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv22_1/relu"
  type: "ReLU"
  bottom: "conv22_1/bn"
  top: "conv22_1/bn"
}
layer {
  name: "conv22_2/dwise"
  type: "Convolution"
  bottom: "conv22_1/bn"
  top: "conv22_2/dwise"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    group: 64
    stride: 2
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv22_2/dwise/bn"
  type: "BatchNorm"
  bottom: "conv22_2/dwise"
  top: "conv22_2/dwise/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv22_2/dwise/scale"
  type: "Scale"
  bottom: "conv22_2/dwise/bn"
  top: "conv22_2/dwise/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv22_2/dwise/relu"
  type: "ReLU"
  bottom: "conv22_2/dwise/bn"
  top: "conv22_2/dwise/bn"
}
layer {
  name: "conv22_3"
  type: "Convolution"
  bottom: "conv22_2/dwise/bn"
  top: "conv22_3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 128
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv22_3/bn"
  type: "BatchNorm"
  bottom: "conv22_3"
  top: "conv22_3/bn"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    use_global_stats: true
    eps: 1e-5
  }
}
layer {
  name: "conv22_3/scale"
  type: "Scale"
  bottom: "conv22_3/bn"
  top: "conv22_3/bn"
  param {
    lr_mult: 1
    decay_mult: 0
  }
  param {
    lr_mult: 1
    decay_mult: 0
  }
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv22_3/relu"
  type: "ReLU"
  bottom: "conv22_3/bn"
  top: "conv22_3/bn"
}
###########################
layer {
  name: "conv14_mbox_loc"
  type: "Convolution"
  bottom: "conv14/expand/bn"
  top: "conv14_mbox_loc"
  param {
    lr_mult: 0.1
    decay_mult: 0.1
  }
  param {
    lr_mult: 0.2
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 12
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv14_mbox_loc_perm"
  type: "Permute"
  bottom: "conv14_mbox_loc"
  top: "conv14_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv14_mbox_loc_flat"
  type: "Flatten"
  bottom: "conv14_mbox_loc_perm"
  top: "conv14_mbox_loc_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv14_mbox_conf_new"
  type: "Convolution"
  bottom: "conv14/expand/bn"
  top: "conv14_mbox_conf"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 63
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv14_mbox_conf_perm"
  type: "Permute"
  bottom: "conv14_mbox_conf"
  top: "conv14_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv14_mbox_conf_flat"
  type: "Flatten"
  bottom: "conv14_mbox_conf_perm"
  top: "conv14_mbox_conf_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv14_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv14/expand/bn"
  bottom: "data"
  top: "conv14_mbox_priorbox"
  prior_box_param {
    min_size: 60.0
    aspect_ratio: 2.0
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    offset: 0.5
  }
}
layer {
  name: "conv18_mbox_loc"
  type: "Convolution"
  bottom: "conv18/bn"
  top: "conv18_mbox_loc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 24
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv18_mbox_loc_perm"
  type: "Permute"
  bottom: "conv18_mbox_loc"
  top: "conv18_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv18_mbox_loc_flat"
  type: "Flatten"
  bottom: "conv18_mbox_loc_perm"
  top: "conv18_mbox_loc_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv18_mbox_conf_new"
  type: "Convolution"
  bottom: "conv18/bn"
  top: "conv18_mbox_conf"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 126
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv18_mbox_conf_perm"
  type: "Permute"
  bottom: "conv18_mbox_conf"
  top: "conv18_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv18_mbox_conf_flat"
  type: "Flatten"
  bottom: "conv18_mbox_conf_perm"
  top: "conv18_mbox_conf_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv18_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv18/bn"
  bottom: "data"
  top: "conv18_mbox_priorbox"
  prior_box_param {
    min_size: 105.0
    max_size: 150.0
    aspect_ratio: 2.0
    aspect_ratio: 3.0
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    offset: 0.5
  }
}
layer {
  name: "conv19_3_mbox_loc"
  type: "Convolution"
  bottom: "conv19_3/bn"
  top: "conv19_3_mbox_loc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 24
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv19_3_mbox_loc_perm"
  type: "Permute"
  bottom: "conv19_3_mbox_loc"
  top: "conv19_3_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv19_3_mbox_loc_flat"
  type: "Flatten"
  bottom: "conv19_3_mbox_loc_perm"
  top: "conv19_3_mbox_loc_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv19_3_mbox_conf_new"
  type: "Convolution"
  bottom: "conv19_3/bn"
  top: "conv19_3_mbox_conf"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 126
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv19_3_mbox_conf_perm"
  type: "Permute"
  bottom: "conv19_3_mbox_conf"
  top: "conv19_3_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv19_3_mbox_conf_flat"
  type: "Flatten"
  bottom: "conv19_3_mbox_conf_perm"
  top: "conv19_3_mbox_conf_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv19_3_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv19_3/bn"
  bottom: "data"
  top: "conv19_3_mbox_priorbox"
  prior_box_param {
    min_size: 150.0
    max_size: 195.0
    aspect_ratio: 2.0
    aspect_ratio: 3.0
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    offset: 0.5
  }
}
layer {
  name: "conv20_3_mbox_loc"
  type: "Convolution"
  bottom: "conv20_3/bn"
  top: "conv20_3_mbox_loc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 24
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv20_3_mbox_loc_perm"
  type: "Permute"
  bottom: "conv20_3_mbox_loc"
  top: "conv20_3_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv20_3_mbox_loc_flat"
  type: "Flatten"
  bottom: "conv20_3_mbox_loc_perm"
  top: "conv20_3_mbox_loc_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv20_3_mbox_conf_new"
  type: "Convolution"
  bottom: "conv20_3/bn"
  top: "conv20_3_mbox_conf"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 126
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv20_3_mbox_conf_perm"
  type: "Permute"
  bottom: "conv20_3_mbox_conf"
  top: "conv20_3_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv20_3_mbox_conf_flat"
  type: "Flatten"
  bottom: "conv20_3_mbox_conf_perm"
  top: "conv20_3_mbox_conf_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv20_3_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv20_3/bn"
  bottom: "data"
  top: "conv20_3_mbox_priorbox"
  prior_box_param {
    min_size: 195.0
    max_size: 240.0
    aspect_ratio: 2.0
    aspect_ratio: 3.0
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    offset: 0.5
  }
}
layer {
  name: "conv21_3_mbox_loc"
  type: "Convolution"
  bottom: "conv21_3/bn"
  top: "conv21_3_mbox_loc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 24
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv21_3_mbox_loc_perm"
  type: "Permute"
  bottom: "conv21_3_mbox_loc"
  top: "conv21_3_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv21_3_mbox_loc_flat"
  type: "Flatten"
  bottom: "conv21_3_mbox_loc_perm"
  top: "conv21_3_mbox_loc_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv21_3_mbox_conf_new"
  type: "Convolution"
  bottom: "conv21_3/bn"
  top: "conv21_3_mbox_conf"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 126
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv21_3_mbox_conf_perm"
  type: "Permute"
  bottom: "conv21_3_mbox_conf"
  top: "conv21_3_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv21_3_mbox_conf_flat"
  type: "Flatten"
  bottom: "conv21_3_mbox_conf_perm"
  top: "conv21_3_mbox_conf_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv21_3_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv21_3/bn"
  bottom: "data"
  top: "conv21_3_mbox_priorbox"
  prior_box_param {
    min_size: 240.0
    max_size: 285.0
    aspect_ratio: 2.0
    aspect_ratio: 3.0
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    offset: 0.5
  }
}
layer {
  name: "conv22_3_mbox_loc"
  type: "Convolution"
  bottom: "conv22_3/bn"
  top: "conv22_3_mbox_loc"
  param {
    lr_mult: 0.1
    decay_mult: 0.1
  }
  param {
    lr_mult: 0.2
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 24
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv22_3_mbox_loc_perm"
  type: "Permute"
  bottom: "conv22_3_mbox_loc"
  top: "conv22_3_mbox_loc_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv22_3_mbox_loc_flat"
  type: "Flatten"
  bottom: "conv22_3_mbox_loc_perm"
  top: "conv22_3_mbox_loc_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv22_3_mbox_conf_new"
  type: "Convolution"
  bottom: "conv22_3/bn"
  top: "conv22_3_mbox_conf"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    pad: 1
    num_output: 126
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv22_3_mbox_conf_perm"
  type: "Permute"
  bottom: "conv22_3_mbox_conf"
  top: "conv22_3_mbox_conf_perm"
  permute_param {
    order: 0
    order: 2
    order: 3
    order: 1
  }
}
layer {
  name: "conv22_3_mbox_conf_flat"
  type: "Flatten"
  bottom: "conv22_3_mbox_conf_perm"
  top: "conv22_3_mbox_conf_flat"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "conv22_3_mbox_priorbox"
  type: "PriorBox"
  bottom: "conv22_3/bn"
  bottom: "data"
  top: "conv22_3_mbox_priorbox"
  prior_box_param {
    min_size: 285.0
    max_size: 300.0
    aspect_ratio: 2.0
    aspect_ratio: 3.0
    flip: true
    clip: false
    variance: 0.1
    variance: 0.1
    variance: 0.2
    variance: 0.2
    offset: 0.5
  }
}
layer {
  name: "mbox_loc"
  type: "Concat"
  bottom: "conv14_mbox_loc_flat"
  bottom: "conv18_mbox_loc_flat"
  bottom: "conv19_3_mbox_loc_flat"
  bottom: "conv20_3_mbox_loc_flat"
  bottom: "conv21_3_mbox_loc_flat"
  bottom: "conv22_3_mbox_loc_flat"
  top: "mbox_loc"
  concat_param {
    axis: 1
  }
}
layer {
  name: "mbox_conf"
  type: "Concat"
  bottom: "conv14_mbox_conf_flat"
  bottom: "conv18_mbox_conf_flat"
  bottom: "conv19_3_mbox_conf_flat"
  bottom: "conv20_3_mbox_conf_flat"
  bottom: "conv21_3_mbox_conf_flat"
  bottom: "conv22_3_mbox_conf_flat"
  top: "mbox_conf"
  concat_param {
    axis: 1
  }
}
layer {
  name: "mbox_priorbox"
  type: "Concat"
  bottom: "conv14_mbox_priorbox"
  bottom: "conv18_mbox_priorbox"
  bottom: "conv19_3_mbox_priorbox"
  bottom: "conv20_3_mbox_priorbox"
  bottom: "conv21_3_mbox_priorbox"
  bottom: "conv22_3_mbox_priorbox"
  top: "mbox_priorbox"
  concat_param {
    axis: 2
  }
}
layer {
  name: "mbox_conf_reshape"
  type: "Reshape"
  bottom: "mbox_conf"
  top: "mbox_conf_reshape"
  reshape_param {
    shape {
      dim: 0
      dim: -1
      dim: 21
    }
  }
}
layer {
  name: "mbox_conf_softmax"
  type: "Softmax"
  bottom: "mbox_conf_reshape"
  top: "mbox_conf_softmax"
  softmax_param {
    axis: 2
  }
}
layer {
  name: "mbox_conf_flatten"
  type: "Flatten"
  bottom: "mbox_conf_softmax"
  top: "mbox_conf_flatten"
  flatten_param {
    axis: 1
  }
}
layer {
  name: "detection_out"
  type: "DetectionOutput"
  bottom: "mbox_loc"
  bottom: "mbox_conf_flatten"
  bottom: "mbox_priorbox"
  top: "detection_out"
  include {
    phase: TEST
  }
  detection_output_param {
    num_classes: 21
    share_location: true
    background_label_id: 0
    nms_param {
      nms_threshold: 0.45
      top_k: 100
    }
    code_type: CENTER_SIZE
    keep_top_k: 100
    confidence_threshold: 0.25
  }
}