mobilenet_v2_deploy.prototxt

name: "mobilenet_v2"
input: "data"
input_shape { 
  dim: 1
  dim: 3
  dim: 224
  dim: 224
}
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 2
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv1_bn"
  type: "BatchNorm"
  bottom: "conv1"
  top: "conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv1_scale"
  bottom: "conv1"
  top: "conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv1_relu"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "resx1_conv1"
  type: "Convolution"
  bottom: "conv1"
  top: "resx1_conv1"
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx1_conv1_bn"
  type: "BatchNorm"
  bottom: "resx1_conv1"
  top: "resx1_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx1_conv1_scale"
  bottom: "resx1_conv1"
  top: "resx1_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx1_conv1_relu"
  type: "ReLU"
  bottom: "resx1_conv1"
  top: "resx1_conv1"
}
layer {
  name: "resx1_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx1_conv1"
  top: "resx1_conv2"
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx1_conv2_bn"
  type: "BatchNorm"
  bottom: "resx1_conv2"
  top: "resx1_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx1_conv2_scale"
  bottom: "resx1_conv2"
  top: "resx1_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx1_conv2_relu"
  type: "ReLU"
  bottom: "resx1_conv2"
  top: "resx1_conv2"
}
layer {
  name: "resx1_conv3"
  type: "Convolution"
  bottom: "resx1_conv2"
  top: "resx1_conv3"
  convolution_param {
    num_output: 16
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx1_conv3_bn"
  type: "BatchNorm"
  bottom: "resx1_conv3"
  top: "resx1_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx1_conv3_scale"
  bottom: "resx1_conv3"
  top: "resx1_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx2_conv1"
  type: "Convolution"
  bottom: "resx1_conv3"
  top: "resx2_conv1"
  convolution_param {
    num_output: 96
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx2_conv1_bn"
  type: "BatchNorm"
  bottom: "resx2_conv1"
  top: "resx2_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx2_conv1_scale"
  bottom: "resx2_conv1"
  top: "resx2_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx2_conv1_relu"
  type: "ReLU"
  bottom: "resx2_conv1"
  top: "resx2_conv1"
}
layer {
  name: "resx2_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx2_conv1"
  top: "resx2_conv2"
  convolution_param {
    num_output: 96
    kernel_size: 3
    stride: 2
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx2_conv2_bn"
  type: "BatchNorm"
  bottom: "resx2_conv2"
  top: "resx2_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx2_conv2_scale"
  bottom: "resx2_conv2"
  top: "resx2_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx2_conv2_relu"
  type: "ReLU"
  bottom: "resx2_conv2"
  top: "resx2_conv2"
}
layer {
  name: "resx2_conv3"
  type: "Convolution"
  bottom: "resx2_conv2"
  top: "resx2_conv3"
  convolution_param {
    num_output: 24
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx2_conv3_bn"
  type: "BatchNorm"
  bottom: "resx2_conv3"
  top: "resx2_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx2_conv3_scale"
  bottom: "resx2_conv3"
  top: "resx2_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx3_conv1"
  type: "Convolution"
  bottom: "resx2_conv3"
  top: "resx3_conv1"
  convolution_param {
    num_output: 144
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx3_conv1_bn"
  type: "BatchNorm"
  bottom: "resx3_conv1"
  top: "resx3_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx3_conv1_scale"
  bottom: "resx3_conv1"
  top: "resx3_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx3_conv1_relu"
  type: "ReLU"
  bottom: "resx3_conv1"
  top: "resx3_conv1"
}
layer {
  name: "resx3_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx3_conv1"
  top: "resx3_conv2"
  convolution_param {
    num_output: 144
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx3_conv2_bn"
  type: "BatchNorm"
  bottom: "resx3_conv2"
  top: "resx3_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx3_conv2_scale"
  bottom: "resx3_conv2"
  top: "resx3_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx3_conv2_relu"
  type: "ReLU"
  bottom: "resx3_conv2"
  top: "resx3_conv2"
}
layer {
  name: "resx3_conv3"
  type: "Convolution"
  bottom: "resx3_conv2"
  top: "resx3_conv3"
  convolution_param {
    num_output: 24
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx3_conv3_bn"
  type: "BatchNorm"
  bottom: "resx3_conv3"
  top: "resx3_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx3_conv3_scale"
  bottom: "resx3_conv3"
  top: "resx3_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx3_add"
  type: "Eltwise"
  bottom: "resx2_conv3"
  bottom: "resx3_conv3"
  top: "resx3_add"
}
layer {
  name: "resx4_conv1"
  type: "Convolution"
  bottom: "resx3_add"
  top: "resx4_conv1"
  convolution_param {
    num_output: 144
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx4_conv1_bn"
  type: "BatchNorm"
  bottom: "resx4_conv1"
  top: "resx4_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx4_conv1_scale"
  bottom: "resx4_conv1"
  top: "resx4_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx4_conv1_relu"
  type: "ReLU"
  bottom: "resx4_conv1"
  top: "resx4_conv1"
}
layer {
  name: "resx4_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx4_conv1"
  top: "resx4_conv2"
  convolution_param {
    num_output: 144
    kernel_size: 3
    stride: 2
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx4_conv2_bn"
  type: "BatchNorm"
  bottom: "resx4_conv2"
  top: "resx4_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx4_conv2_scale"
  bottom: "resx4_conv2"
  top: "resx4_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx4_conv2_relu"
  type: "ReLU"
  bottom: "resx4_conv2"
  top: "resx4_conv2"
}
layer {
  name: "resx4_conv3"
  type: "Convolution"
  bottom: "resx4_conv2"
  top: "resx4_conv3"
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx4_conv3_bn"
  type: "BatchNorm"
  bottom: "resx4_conv3"
  top: "resx4_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx4_conv3_scale"
  bottom: "resx4_conv3"
  top: "resx4_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx5_conv1"
  type: "Convolution"
  bottom: "resx4_conv3"
  top: "resx5_conv1"
  convolution_param {
    num_output: 192
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx5_conv1_bn"
  type: "BatchNorm"
  bottom: "resx5_conv1"
  top: "resx5_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx5_conv1_scale"
  bottom: "resx5_conv1"
  top: "resx5_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx5_conv1_relu"
  type: "ReLU"
  bottom: "resx5_conv1"
  top: "resx5_conv1"
}
layer {
  name: "resx5_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx5_conv1"
  top: "resx5_conv2"
  convolution_param {
    num_output: 192
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx5_conv2_bn"
  type: "BatchNorm"
  bottom: "resx5_conv2"
  top: "resx5_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx5_conv2_scale"
  bottom: "resx5_conv2"
  top: "resx5_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx5_conv2_relu"
  type: "ReLU"
  bottom: "resx5_conv2"
  top: "resx5_conv2"
}
layer {
  name: "resx5_conv3"
  type: "Convolution"
  bottom: "resx5_conv2"
  top: "resx5_conv3"
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx5_conv3_bn"
  type: "BatchNorm"
  bottom: "resx5_conv3"
  top: "resx5_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx5_conv3_scale"
  bottom: "resx5_conv3"
  top: "resx5_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx5_add"
  type: "Eltwise"
  bottom: "resx4_conv3"
  bottom: "resx5_conv3"
  top: "resx5_add"
}
layer {
  name: "resx6_conv1"
  type: "Convolution"
  bottom: "resx5_add"
  top: "resx6_conv1"
  convolution_param {
    num_output: 192
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx6_conv1_bn"
  type: "BatchNorm"
  bottom: "resx6_conv1"
  top: "resx6_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx6_conv1_scale"
  bottom: "resx6_conv1"
  top: "resx6_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx6_conv1_relu"
  type: "ReLU"
  bottom: "resx6_conv1"
  top: "resx6_conv1"
}
layer {
  name: "resx6_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx6_conv1"
  top: "resx6_conv2"
  convolution_param {
    num_output: 192
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx6_conv2_bn"
  type: "BatchNorm"
  bottom: "resx6_conv2"
  top: "resx6_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx6_conv2_scale"
  bottom: "resx6_conv2"
  top: "resx6_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx6_conv2_relu"
  type: "ReLU"
  bottom: "resx6_conv2"
  top: "resx6_conv2"
}
layer {
  name: "resx6_conv3"
  type: "Convolution"
  bottom: "resx6_conv2"
  top: "resx6_conv3"
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx6_conv3_bn"
  type: "BatchNorm"
  bottom: "resx6_conv3"
  top: "resx6_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx6_conv3_scale"
  bottom: "resx6_conv3"
  top: "resx6_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx6_add"
  type: "Eltwise"
  bottom: "resx5_add"
  bottom: "resx6_conv3"
  top: "resx6_add"
}
layer {
  name: "resx7_conv1"
  type: "Convolution"
  bottom: "resx6_add"
  top: "resx7_conv1"
  convolution_param {
    num_output: 192
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx7_conv1_bn"
  type: "BatchNorm"
  bottom: "resx7_conv1"
  top: "resx7_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx7_conv1_scale"
  bottom: "resx7_conv1"
  top: "resx7_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx7_conv1_relu"
  type: "ReLU"
  bottom: "resx7_conv1"
  top: "resx7_conv1"
}
layer {
  name: "resx7_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx7_conv1"
  top: "resx7_conv2"
  convolution_param {
    num_output: 192
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx7_conv2_bn"
  type: "BatchNorm"
  bottom: "resx7_conv2"
  top: "resx7_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx7_conv2_scale"
  bottom: "resx7_conv2"
  top: "resx7_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx7_conv2_relu"
  type: "ReLU"
  bottom: "resx7_conv2"
  top: "resx7_conv2"
}
layer {
  name: "resx7_conv3"
  type: "Convolution"
  bottom: "resx7_conv2"
  top: "resx7_conv3"
  convolution_param {
    num_output: 64
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx7_conv3_bn"
  type: "BatchNorm"
  bottom: "resx7_conv3"
  top: "resx7_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx7_conv3_scale"
  bottom: "resx7_conv3"
  top: "resx7_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx8_conv1"
  type: "Convolution"
  bottom: "resx7_conv3"
  top: "resx8_conv1"
  convolution_param {
    num_output: 384
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx8_conv1_bn"
  type: "BatchNorm"
  bottom: "resx8_conv1"
  top: "resx8_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx8_conv1_scale"
  bottom: "resx8_conv1"
  top: "resx8_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx8_conv1_relu"
  type: "ReLU"
  bottom: "resx8_conv1"
  top: "resx8_conv1"
}
layer {
  name: "resx8_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx8_conv1"
  top: "resx8_conv2"
  convolution_param {
    num_output: 384
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx8_conv2_bn"
  type: "BatchNorm"
  bottom: "resx8_conv2"
  top: "resx8_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx8_conv2_scale"
  bottom: "resx8_conv2"
  top: "resx8_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx8_conv2_relu"
  type: "ReLU"
  bottom: "resx8_conv2"
  top: "resx8_conv2"
}
layer {
  name: "resx8_conv3"
  type: "Convolution"
  bottom: "resx8_conv2"
  top: "resx8_conv3"
  convolution_param {
    num_output: 64
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx8_conv3_bn"
  type: "BatchNorm"
  bottom: "resx8_conv3"
  top: "resx8_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx8_conv3_scale"
  bottom: "resx8_conv3"
  top: "resx8_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx8_add"
  type: "Eltwise"
  bottom: "resx7_conv3"
  bottom: "resx8_conv3"
  top: "resx8_add"
}
layer {
  name: "resx9_conv1"
  type: "Convolution"
  bottom: "resx8_add"
  top: "resx9_conv1"
  convolution_param {
    num_output: 384
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx9_conv1_bn"
  type: "BatchNorm"
  bottom: "resx9_conv1"
  top: "resx9_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx9_conv1_scale"
  bottom: "resx9_conv1"
  top: "resx9_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx9_conv1_relu"
  type: "ReLU"
  bottom: "resx9_conv1"
  top: "resx9_conv1"
}
layer {
  name: "resx9_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx9_conv1"
  top: "resx9_conv2"
  convolution_param {
    num_output: 384
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx9_conv2_bn"
  type: "BatchNorm"
  bottom: "resx9_conv2"
  top: "resx9_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx9_conv2_scale"
  bottom: "resx9_conv2"
  top: "resx9_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx9_conv2_relu"
  type: "ReLU"
  bottom: "resx9_conv2"
  top: "resx9_conv2"
}
layer {
  name: "resx9_conv3"
  type: "Convolution"
  bottom: "resx9_conv2"
  top: "resx9_conv3"
  convolution_param {
    num_output: 64
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx9_conv3_bn"
  type: "BatchNorm"
  bottom: "resx9_conv3"
  top: "resx9_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx9_conv3_scale"
  bottom: "resx9_conv3"
  top: "resx9_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx9_add"
  type: "Eltwise"
  bottom: "resx8_add"
  bottom: "resx9_conv3"
  top: "resx9_add"
}
layer {
  name: "resx10_conv1"
  type: "Convolution"
  bottom: "resx9_add"
  top: "resx10_conv1"
  convolution_param {
    num_output: 384
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx10_conv1_bn"
  type: "BatchNorm"
  bottom: "resx10_conv1"
  top: "resx10_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx10_conv1_scale"
  bottom: "resx10_conv1"
  top: "resx10_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx10_conv1_relu"
  type: "ReLU"
  bottom: "resx10_conv1"
  top: "resx10_conv1"
}
layer {
  name: "resx10_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx10_conv1"
  top: "resx10_conv2"
  convolution_param {
    num_output: 384
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx10_conv2_bn"
  type: "BatchNorm"
  bottom: "resx10_conv2"
  top: "resx10_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx10_conv2_scale"
  bottom: "resx10_conv2"
  top: "resx10_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx10_conv2_relu"
  type: "ReLU"
  bottom: "resx10_conv2"
  top: "resx10_conv2"
}
layer {
  name: "resx10_conv3"
  type: "Convolution"
  bottom: "resx10_conv2"
  top: "resx10_conv3"
  convolution_param {
    num_output: 64
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx10_conv3_bn"
  type: "BatchNorm"
  bottom: "resx10_conv3"
  top: "resx10_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx10_conv3_scale"
  bottom: "resx10_conv3"
  top: "resx10_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx10_add"
  type: "Eltwise"
  bottom: "resx9_add"
  bottom: "resx10_conv3"
  top: "resx10_add"
}
layer {
  name: "resx11_conv1"
  type: "Convolution"
  bottom: "resx10_add"
  top: "resx11_conv1"
  convolution_param {
    num_output: 384
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx11_conv1_bn"
  type: "BatchNorm"
  bottom: "resx11_conv1"
  top: "resx11_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx11_conv1_scale"
  bottom: "resx11_conv1"
  top: "resx11_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx11_conv1_relu"
  type: "ReLU"
  bottom: "resx11_conv1"
  top: "resx11_conv1"
}
layer {
  name: "resx11_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx11_conv1"
  top: "resx11_conv2"
  convolution_param {
    num_output: 384
    kernel_size: 3
    stride: 2
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx11_conv2_bn"
  type: "BatchNorm"
  bottom: "resx11_conv2"
  top: "resx11_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx11_conv2_scale"
  bottom: "resx11_conv2"
  top: "resx11_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx11_conv2_relu"
  type: "ReLU"
  bottom: "resx11_conv2"
  top: "resx11_conv2"
}
layer {
  name: "resx11_conv3"
  type: "Convolution"
  bottom: "resx11_conv2"
  top: "resx11_conv3"
  convolution_param {
    num_output: 96
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx11_conv3_bn"
  type: "BatchNorm"
  bottom: "resx11_conv3"
  top: "resx11_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx11_conv3_scale"
  bottom: "resx11_conv3"
  top: "resx11_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx12_conv1"
  type: "Convolution"
  bottom: "resx11_conv3"
  top: "resx12_conv1"
  convolution_param {
    num_output: 576
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx12_conv1_bn"
  type: "BatchNorm"
  bottom: "resx12_conv1"
  top: "resx12_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx12_conv1_scale"
  bottom: "resx12_conv1"
  top: "resx12_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx12_conv1_relu"
  type: "ReLU"
  bottom: "resx12_conv1"
  top: "resx12_conv1"
}
layer {
  name: "resx12_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx12_conv1"
  top: "resx12_conv2"
  convolution_param {
    num_output: 576
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx12_conv2_bn"
  type: "BatchNorm"
  bottom: "resx12_conv2"
  top: "resx12_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx12_conv2_scale"
  bottom: "resx12_conv2"
  top: "resx12_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx12_conv2_relu"
  type: "ReLU"
  bottom: "resx12_conv2"
  top: "resx12_conv2"
}
layer {
  name: "resx12_conv3"
  type: "Convolution"
  bottom: "resx12_conv2"
  top: "resx12_conv3"
  convolution_param {
    num_output: 96
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx12_conv3_bn"
  type: "BatchNorm"
  bottom: "resx12_conv3"
  top: "resx12_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx12_conv3_scale"
  bottom: "resx12_conv3"
  top: "resx12_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx12_add"
  type: "Eltwise"
  bottom: "resx11_conv3"
  bottom: "resx12_conv3"
  top: "resx12_add"
}
layer {
  name: "resx13_conv1"
  type: "Convolution"
  bottom: "resx12_add"
  top: "resx13_conv1"
  convolution_param {
    num_output: 576
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx13_conv1_bn"
  type: "BatchNorm"
  bottom: "resx13_conv1"
  top: "resx13_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx13_conv1_scale"
  bottom: "resx13_conv1"
  top: "resx13_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx13_conv1_relu"
  type: "ReLU"
  bottom: "resx13_conv1"
  top: "resx13_conv1"
}
layer {
  name: "resx13_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx13_conv1"
  top: "resx13_conv2"
  convolution_param {
    num_output: 576
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx13_conv2_bn"
  type: "BatchNorm"
  bottom: "resx13_conv2"
  top: "resx13_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx13_conv2_scale"
  bottom: "resx13_conv2"
  top: "resx13_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx13_conv2_relu"
  type: "ReLU"
  bottom: "resx13_conv2"
  top: "resx13_conv2"
}
layer {
  name: "resx13_conv3"
  type: "Convolution"
  bottom: "resx13_conv2"
  top: "resx13_conv3"
  convolution_param {
    num_output: 96
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx13_conv3_bn"
  type: "BatchNorm"
  bottom: "resx13_conv3"
  top: "resx13_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx13_conv3_scale"
  bottom: "resx13_conv3"
  top: "resx13_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx13_add"
  type: "Eltwise"
  bottom: "resx12_add"
  bottom: "resx13_conv3"
  top: "resx13_add"
}
layer {
  name: "resx14_conv1"
  type: "Convolution"
  bottom: "resx13_add"
  top: "resx14_conv1"
  convolution_param {
    num_output: 576
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx14_conv1_bn"
  type: "BatchNorm"
  bottom: "resx14_conv1"
  top: "resx14_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx14_conv1_scale"
  bottom: "resx14_conv1"
  top: "resx14_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx14_conv1_relu"
  type: "ReLU"
  bottom: "resx14_conv1"
  top: "resx14_conv1"
}
layer {
  name: "resx14_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx14_conv1"
  top: "resx14_conv2"
  convolution_param {
    num_output: 576
    kernel_size: 3
    stride: 2
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx14_conv2_bn"
  type: "BatchNorm"
  bottom: "resx14_conv2"
  top: "resx14_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx14_conv2_scale"
  bottom: "resx14_conv2"
  top: "resx14_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx14_conv2_relu"
  type: "ReLU"
  bottom: "resx14_conv2"
  top: "resx14_conv2"
}
layer {
  name: "resx14_conv3"
  type: "Convolution"
  bottom: "resx14_conv2"
  top: "resx14_conv3"
  convolution_param {
    num_output: 160
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx14_conv3_bn"
  type: "BatchNorm"
  bottom: "resx14_conv3"
  top: "resx14_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx14_conv3_scale"
  bottom: "resx14_conv3"
  top: "resx14_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx15_conv1"
  type: "Convolution"
  bottom: "resx14_conv3"
  top: "resx15_conv1"
  convolution_param {
    num_output: 960
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx15_conv1_bn"
  type: "BatchNorm"
  bottom: "resx15_conv1"
  top: "resx15_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx15_conv1_scale"
  bottom: "resx15_conv1"
  top: "resx15_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx15_conv1_relu"
  type: "ReLU"
  bottom: "resx15_conv1"
  top: "resx15_conv1"
}
layer {
  name: "resx15_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx15_conv1"
  top: "resx15_conv2"
  convolution_param {
    num_output: 960
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx15_conv2_bn"
  type: "BatchNorm"
  bottom: "resx15_conv2"
  top: "resx15_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx15_conv2_scale"
  bottom: "resx15_conv2"
  top: "resx15_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx15_conv2_relu"
  type: "ReLU"
  bottom: "resx15_conv2"
  top: "resx15_conv2"
}
layer {
  name: "resx15_conv3"
  type: "Convolution"
  bottom: "resx15_conv2"
  top: "resx15_conv3"
  convolution_param {
    num_output: 160
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx15_conv3_bn"
  type: "BatchNorm"
  bottom: "resx15_conv3"
  top: "resx15_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx15_conv3_scale"
  bottom: "resx15_conv3"
  top: "resx15_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx15_add"
  type: "Eltwise"
  bottom: "resx14_conv3"
  bottom: "resx15_conv3"
  top: "resx15_add"
}
layer {
  name: "resx16_conv1"
  type: "Convolution"
  bottom: "resx15_add"
  top: "resx16_conv1"
  convolution_param {
    num_output: 960
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx16_conv1_bn"
  type: "BatchNorm"
  bottom: "resx16_conv1"
  top: "resx16_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx16_conv1_scale"
  bottom: "resx16_conv1"
  top: "resx16_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx16_conv1_relu"
  type: "ReLU"
  bottom: "resx16_conv1"
  top: "resx16_conv1"
}
layer {
  name: "resx16_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx16_conv1"
  top: "resx16_conv2"
  convolution_param {
    num_output: 960
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx16_conv2_bn"
  type: "BatchNorm"
  bottom: "resx16_conv2"
  top: "resx16_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx16_conv2_scale"
  bottom: "resx16_conv2"
  top: "resx16_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx16_conv2_relu"
  type: "ReLU"
  bottom: "resx16_conv2"
  top: "resx16_conv2"
}
layer {
  name: "resx16_conv3"
  type: "Convolution"
  bottom: "resx16_conv2"
  top: "resx16_conv3"
  convolution_param {
    num_output: 160
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx16_conv3_bn"
  type: "BatchNorm"
  bottom: "resx16_conv3"
  top: "resx16_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx16_conv3_scale"
  bottom: "resx16_conv3"
  top: "resx16_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx16_add"
  type: "Eltwise"
  bottom: "resx15_add"
  bottom: "resx16_conv3"
  top: "resx16_add"
}
layer {
  name: "resx17_conv1"
  type: "Convolution"
  bottom: "resx16_add"
  top: "resx17_conv1"
  convolution_param {
    num_output: 960
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx17_conv1_bn"
  type: "BatchNorm"
  bottom: "resx17_conv1"
  top: "resx17_conv1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx17_conv1_scale"
  bottom: "resx17_conv1"
  top: "resx17_conv1"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx17_conv1_relu"
  type: "ReLU"
  bottom: "resx17_conv1"
  top: "resx17_conv1"
}
layer {
  name: "resx17_conv2"
  type: "ConvolutionDepthwise"
  bottom: "resx17_conv1"
  top: "resx17_conv2"
  convolution_param {
    num_output: 960
    kernel_size: 3
    stride: 1
    pad: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx17_conv2_bn"
  type: "BatchNorm"
  bottom: "resx17_conv2"
  top: "resx17_conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx17_conv2_scale"
  bottom: "resx17_conv2"
  top: "resx17_conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "resx17_conv2_relu"
  type: "ReLU"
  bottom: "resx17_conv2"
  top: "resx17_conv2"
}
layer {
  name: "resx17_conv3"
  type: "Convolution"
  bottom: "resx17_conv2"
  top: "resx17_conv3"
  convolution_param {
    num_output: 320
    kernel_size: 1
    stride: 1
    pad: 0
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "resx17_conv3_bn"
  type: "BatchNorm"
  bottom: "resx17_conv3"
  top: "resx17_conv3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "resx17_conv3_scale"
  bottom: "resx17_conv3"
  top: "resx17_conv3"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "resx17_conv3"
  top: "conv2"
  convolution_param {
    num_output: 1280
    pad: 0
    kernel_size: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_bn"
  type: "BatchNorm"
  bottom: "conv2"
  top: "conv2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
}
layer {
  name: "conv2_scale"
  bottom: "conv2"
  top: "conv2"
  type: "Scale"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "conv2_relu"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}
layer {
  name: "pool_ave"
  type: "Pooling"
  bottom: "conv2"
  top: "pool_ave"
  pooling_param {
    global_pooling : true
    pool: AVE
  }
}
layer {
  name: "fc1000"
  type: "Convolution"
  bottom: "pool_ave"
  top: "fc1000"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 1000
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "loss"
  type: "SoftmaxWithLoss"
  bottom: "fc1000"
  bottom: "label"
  top: "loss"
}