se_resnet_50_v1_deploy.prototxt

name: "SE_RESNET_50_V1"
input: "data"
input_shape { 
  dim: 1
  dim: 3
  dim: 225
  dim: 225
}
#  transform_param {
#    scale: 0.017
#    mirror: false
#    crop_size: 225
#    mean_value: [103.94,116.78,123.68]
#  }
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 3
    kernel_size: 7
    stride: 2
    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
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv1/scale"
  type: "Scale"
  bottom: "conv1"
  top: "conv1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}
layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv2_1/prj"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2_1/prj"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_1/prj/bn"
  type: "BatchNorm"
  bottom: "conv2_1/prj"
  top: "conv2_1/prj"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_1/prj/scale"
  type: "Scale"
  bottom: "conv2_1/prj"
  top: "conv2_1/prj"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv2_1/x1"
  type: "Convolution"
  bottom: "pool1"
  top: "conv2_1/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_1/x1/bn"
  type: "BatchNorm"
  bottom: "conv2_1/x1"
  top: "conv2_1/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_1/x1/scale"
  type: "Scale"
  bottom: "conv2_1/x1"
  top: "conv2_1/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_1/x1"
  type: "ReLU"
  bottom: "conv2_1/x1"
  top: "conv2_1/x1"
}
layer {
  name: "conv2_1/x2"
  type: "Convolution"
  bottom: "conv2_1/x1"
  top: "conv2_1/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_1/x2/bn"
  type: "BatchNorm"
  bottom: "conv2_1/x2"
  top: "conv2_1/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_1/x2/scale"
  type: "Scale"
  bottom: "conv2_1/x2"
  top: "conv2_1/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_1/x2"
  type: "ReLU"
  bottom: "conv2_1/x2"
  top: "conv2_1/x2"
}
layer {
  name: "conv2_1/x3"
  type: "Convolution"
  bottom: "conv2_1/x2"
  top: "conv2_1/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_1/x3/bn"
  type: "BatchNorm"
  bottom: "conv2_1/x3"
  top: "conv2_1/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_1/x3/scale"
  type: "Scale"
  bottom: "conv2_1/x3"
  top: "conv2_1/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool2_1/gap"
  type: "Pooling"
  bottom: "conv2_1/x3"
  top: "pool2_1/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc2_1/sqz"
  type: "InnerProduct"
  bottom: "pool2_1/gap"
  top: "fc2_1/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 16
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu2_1/sqz"
  type: "ReLU"
  bottom: "fc2_1/sqz"
  top: "fc2_1/sqz"
}
layer {
  name: "fc2_1/exc"
  type: "InnerProduct"
  bottom: "fc2_1/sqz"
  top: "fc2_1/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 256
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm2_1/gate"
  type: "Sigmoid"
  bottom: "fc2_1/exc"
  top: "fc2_1/exc"
}
layer {
  name: "scale2_1"
  type: "Scale"
  bottom: "conv2_1/x3"
  bottom: "fc2_1/exc"
  top: "scale2_1"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_2_1"
  type: "Eltwise"
  bottom: "conv2_1/prj"
  bottom: "scale2_1"
  top: "block_2_1"
}
layer {
  name: "relu2_1"
  type: "ReLU"
  bottom: "block_2_1"
  top: "block_2_1"
}
layer {
  name: "conv2_2/x1"
  type: "Convolution"
  bottom: "block_2_1"
  top: "conv2_2/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_2/x1/bn"
  type: "BatchNorm"
  bottom: "conv2_2/x1"
  top: "conv2_2/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_2/x1/scale"
  type: "Scale"
  bottom: "conv2_2/x1"
  top: "conv2_2/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_2/x1"
  type: "ReLU"
  bottom: "conv2_2/x1"
  top: "conv2_2/x1"
}
layer {
  name: "conv2_2/x2"
  type: "Convolution"
  bottom: "conv2_2/x1"
  top: "conv2_2/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_2/x2/bn"
  type: "BatchNorm"
  bottom: "conv2_2/x2"
  top: "conv2_2/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_2/x2/scale"
  type: "Scale"
  bottom: "conv2_2/x2"
  top: "conv2_2/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_2/x2"
  type: "ReLU"
  bottom: "conv2_2/x2"
  top: "conv2_2/x2"
}
layer {
  name: "conv2_2/x3"
  type: "Convolution"
  bottom: "conv2_2/x2"
  top: "conv2_2/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_2/x3/bn"
  type: "BatchNorm"
  bottom: "conv2_2/x3"
  top: "conv2_2/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_2/x3/scale"
  type: "Scale"
  bottom: "conv2_2/x3"
  top: "conv2_2/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool2_2/gap"
  type: "Pooling"
  bottom: "conv2_2/x3"
  top: "pool2_2/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc2_2/sqz"
  type: "InnerProduct"
  bottom: "pool2_2/gap"
  top: "fc2_2/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 16
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu2_2/sqz"
  type: "ReLU"
  bottom: "fc2_2/sqz"
  top: "fc2_2/sqz"
}
layer {
  name: "fc2_2/exc"
  type: "InnerProduct"
  bottom: "fc2_2/sqz"
  top: "fc2_2/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 256
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm2_2/gate"
  type: "Sigmoid"
  bottom: "fc2_2/exc"
  top: "fc2_2/exc"
}
layer {
  name: "scale2_2"
  type: "Scale"
  bottom: "conv2_2/x3"
  bottom: "fc2_2/exc"
  top: "scale2_2"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_2_2"
  type: "Eltwise"
  bottom: "block_2_1"
  bottom: "scale2_2"
  top: "block_2_2"
}
layer {
  name: "relu2_2"
  type: "ReLU"
  bottom: "block_2_2"
  top: "block_2_2"
}
layer {
  name: "conv2_3/x1"
  type: "Convolution"
  bottom: "block_2_2"
  top: "conv2_3/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_3/x1/bn"
  type: "BatchNorm"
  bottom: "conv2_3/x1"
  top: "conv2_3/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_3/x1/scale"
  type: "Scale"
  bottom: "conv2_3/x1"
  top: "conv2_3/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_3/x1"
  type: "ReLU"
  bottom: "conv2_3/x1"
  top: "conv2_3/x1"
}
layer {
  name: "conv2_3/x2"
  type: "Convolution"
  bottom: "conv2_3/x1"
  top: "conv2_3/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 64
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_3/x2/bn"
  type: "BatchNorm"
  bottom: "conv2_3/x2"
  top: "conv2_3/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_3/x2/scale"
  type: "Scale"
  bottom: "conv2_3/x2"
  top: "conv2_3/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu2_3/x2"
  type: "ReLU"
  bottom: "conv2_3/x2"
  top: "conv2_3/x2"
}
layer {
  name: "conv2_3/x3"
  type: "Convolution"
  bottom: "conv2_3/x2"
  top: "conv2_3/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv2_3/x3/bn"
  type: "BatchNorm"
  bottom: "conv2_3/x3"
  top: "conv2_3/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv2_3/x3/scale"
  type: "Scale"
  bottom: "conv2_3/x3"
  top: "conv2_3/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool2_3/gap"
  type: "Pooling"
  bottom: "conv2_3/x3"
  top: "pool2_3/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc2_3/sqz"
  type: "InnerProduct"
  bottom: "pool2_3/gap"
  top: "fc2_3/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 16
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu2_3/sqz"
  type: "ReLU"
  bottom: "fc2_3/sqz"
  top: "fc2_3/sqz"
}
layer {
  name: "fc2_3/exc"
  type: "InnerProduct"
  bottom: "fc2_3/sqz"
  top: "fc2_3/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 256
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm2_3/gate"
  type: "Sigmoid"
  bottom: "fc2_3/exc"
  top: "fc2_3/exc"
}
layer {
  name: "scale2_3"
  type: "Scale"
  bottom: "conv2_3/x3"
  bottom: "fc2_3/exc"
  top: "scale2_3"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_2_3"
  type: "Eltwise"
  bottom: "block_2_2"
  bottom: "scale2_3"
  top: "block_2_3"
}
layer {
  name: "relu2_3"
  type: "ReLU"
  bottom: "block_2_3"
  top: "block_2_3"
}
layer {
  name: "conv3_1/prj"
  type: "Convolution"
  bottom: "block_2_3"
  top: "conv3_1/prj"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/prj/bn"
  type: "BatchNorm"
  bottom: "conv3_1/prj"
  top: "conv3_1/prj"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_1/prj/scale"
  type: "Scale"
  bottom: "conv3_1/prj"
  top: "conv3_1/prj"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv3_1/x1"
  type: "Convolution"
  bottom: "block_2_3"
  top: "conv3_1/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/x1/bn"
  type: "BatchNorm"
  bottom: "conv3_1/x1"
  top: "conv3_1/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_1/x1/scale"
  type: "Scale"
  bottom: "conv3_1/x1"
  top: "conv3_1/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_1/x1"
  type: "ReLU"
  bottom: "conv3_1/x1"
  top: "conv3_1/x1"
}
layer {
  name: "conv3_1/x2"
  type: "Convolution"
  bottom: "conv3_1/x1"
  top: "conv3_1/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/x2/bn"
  type: "BatchNorm"
  bottom: "conv3_1/x2"
  top: "conv3_1/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_1/x2/scale"
  type: "Scale"
  bottom: "conv3_1/x2"
  top: "conv3_1/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_1/x2"
  type: "ReLU"
  bottom: "conv3_1/x2"
  top: "conv3_1/x2"
}
layer {
  name: "conv3_1/x3"
  type: "Convolution"
  bottom: "conv3_1/x2"
  top: "conv3_1/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_1/x3/bn"
  type: "BatchNorm"
  bottom: "conv3_1/x3"
  top: "conv3_1/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_1/x3/scale"
  type: "Scale"
  bottom: "conv3_1/x3"
  top: "conv3_1/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool3_1/gap"
  type: "Pooling"
  bottom: "conv3_1/x3"
  top: "pool3_1/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc3_1/sqz"
  type: "InnerProduct"
  bottom: "pool3_1/gap"
  top: "fc3_1/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 32
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_1/sqz"
  type: "ReLU"
  bottom: "fc3_1/sqz"
  top: "fc3_1/sqz"
}
layer {
  name: "fc3_1/exc"
  type: "InnerProduct"
  bottom: "fc3_1/sqz"
  top: "fc3_1/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm3_1/gate"
  type: "Sigmoid"
  bottom: "fc3_1/exc"
  top: "fc3_1/exc"
}
layer {
  name: "scale3_1"
  type: "Scale"
  bottom: "conv3_1/x3"
  bottom: "fc3_1/exc"
  top: "scale3_1"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_3_1"
  type: "Eltwise"
  bottom: "conv3_1/prj"
  bottom: "scale3_1"
  top: "block_3_1"
}
layer {
  name: "relu3_1"
  type: "ReLU"
  bottom: "block_3_1"
  top: "block_3_1"
}
layer {
  name: "conv3_2/x1"
  type: "Convolution"
  bottom: "block_3_1"
  top: "conv3_2/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_2/x1/bn"
  type: "BatchNorm"
  bottom: "conv3_2/x1"
  top: "conv3_2/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_2/x1/scale"
  type: "Scale"
  bottom: "conv3_2/x1"
  top: "conv3_2/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_2/x1"
  type: "ReLU"
  bottom: "conv3_2/x1"
  top: "conv3_2/x1"
}
layer {
  name: "conv3_2/x2"
  type: "Convolution"
  bottom: "conv3_2/x1"
  top: "conv3_2/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_2/x2/bn"
  type: "BatchNorm"
  bottom: "conv3_2/x2"
  top: "conv3_2/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_2/x2/scale"
  type: "Scale"
  bottom: "conv3_2/x2"
  top: "conv3_2/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_2/x2"
  type: "ReLU"
  bottom: "conv3_2/x2"
  top: "conv3_2/x2"
}
layer {
  name: "conv3_2/x3"
  type: "Convolution"
  bottom: "conv3_2/x2"
  top: "conv3_2/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_2/x3/bn"
  type: "BatchNorm"
  bottom: "conv3_2/x3"
  top: "conv3_2/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_2/x3/scale"
  type: "Scale"
  bottom: "conv3_2/x3"
  top: "conv3_2/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool3_2/gap"
  type: "Pooling"
  bottom: "conv3_2/x3"
  top: "pool3_2/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc3_2/sqz"
  type: "InnerProduct"
  bottom: "pool3_2/gap"
  top: "fc3_2/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 32
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_2/sqz"
  type: "ReLU"
  bottom: "fc3_2/sqz"
  top: "fc3_2/sqz"
}
layer {
  name: "fc3_2/exc"
  type: "InnerProduct"
  bottom: "fc3_2/sqz"
  top: "fc3_2/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm3_2/gate"
  type: "Sigmoid"
  bottom: "fc3_2/exc"
  top: "fc3_2/exc"
}
layer {
  name: "scale3_2"
  type: "Scale"
  bottom: "conv3_2/x3"
  bottom: "fc3_2/exc"
  top: "scale3_2"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_3_2"
  type: "Eltwise"
  bottom: "block_3_1"
  bottom: "scale3_2"
  top: "block_3_2"
}
layer {
  name: "relu3_2"
  type: "ReLU"
  bottom: "block_3_2"
  top: "block_3_2"
}
layer {
  name: "conv3_3/x1"
  type: "Convolution"
  bottom: "block_3_2"
  top: "conv3_3/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_3/x1/bn"
  type: "BatchNorm"
  bottom: "conv3_3/x1"
  top: "conv3_3/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_3/x1/scale"
  type: "Scale"
  bottom: "conv3_3/x1"
  top: "conv3_3/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_3/x1"
  type: "ReLU"
  bottom: "conv3_3/x1"
  top: "conv3_3/x1"
}
layer {
  name: "conv3_3/x2"
  type: "Convolution"
  bottom: "conv3_3/x1"
  top: "conv3_3/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_3/x2/bn"
  type: "BatchNorm"
  bottom: "conv3_3/x2"
  top: "conv3_3/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_3/x2/scale"
  type: "Scale"
  bottom: "conv3_3/x2"
  top: "conv3_3/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_3/x2"
  type: "ReLU"
  bottom: "conv3_3/x2"
  top: "conv3_3/x2"
}
layer {
  name: "conv3_3/x3"
  type: "Convolution"
  bottom: "conv3_3/x2"
  top: "conv3_3/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_3/x3/bn"
  type: "BatchNorm"
  bottom: "conv3_3/x3"
  top: "conv3_3/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_3/x3/scale"
  type: "Scale"
  bottom: "conv3_3/x3"
  top: "conv3_3/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool3_3/gap"
  type: "Pooling"
  bottom: "conv3_3/x3"
  top: "pool3_3/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc3_3/sqz"
  type: "InnerProduct"
  bottom: "pool3_3/gap"
  top: "fc3_3/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 32
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_3/sqz"
  type: "ReLU"
  bottom: "fc3_3/sqz"
  top: "fc3_3/sqz"
}
layer {
  name: "fc3_3/exc"
  type: "InnerProduct"
  bottom: "fc3_3/sqz"
  top: "fc3_3/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm3_3/gate"
  type: "Sigmoid"
  bottom: "fc3_3/exc"
  top: "fc3_3/exc"
}
layer {
  name: "scale3_3"
  type: "Scale"
  bottom: "conv3_3/x3"
  bottom: "fc3_3/exc"
  top: "scale3_3"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_3_3"
  type: "Eltwise"
  bottom: "block_3_2"
  bottom: "scale3_3"
  top: "block_3_3"
}
layer {
  name: "relu3_3"
  type: "ReLU"
  bottom: "block_3_3"
  top: "block_3_3"
}
layer {
  name: "conv3_4/x1"
  type: "Convolution"
  bottom: "block_3_3"
  top: "conv3_4/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_4/x1/bn"
  type: "BatchNorm"
  bottom: "conv3_4/x1"
  top: "conv3_4/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_4/x1/scale"
  type: "Scale"
  bottom: "conv3_4/x1"
  top: "conv3_4/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_4/x1"
  type: "ReLU"
  bottom: "conv3_4/x1"
  top: "conv3_4/x1"
}
layer {
  name: "conv3_4/x2"
  type: "Convolution"
  bottom: "conv3_4/x1"
  top: "conv3_4/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 128
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_4/x2/bn"
  type: "BatchNorm"
  bottom: "conv3_4/x2"
  top: "conv3_4/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_4/x2/scale"
  type: "Scale"
  bottom: "conv3_4/x2"
  top: "conv3_4/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu3_4/x2"
  type: "ReLU"
  bottom: "conv3_4/x2"
  top: "conv3_4/x2"
}
layer {
  name: "conv3_4/x3"
  type: "Convolution"
  bottom: "conv3_4/x2"
  top: "conv3_4/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3_4/x3/bn"
  type: "BatchNorm"
  bottom: "conv3_4/x3"
  top: "conv3_4/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv3_4/x3/scale"
  type: "Scale"
  bottom: "conv3_4/x3"
  top: "conv3_4/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool3_4/gap"
  type: "Pooling"
  bottom: "conv3_4/x3"
  top: "pool3_4/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc3_4/sqz"
  type: "InnerProduct"
  bottom: "pool3_4/gap"
  top: "fc3_4/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 32
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_4/sqz"
  type: "ReLU"
  bottom: "fc3_4/sqz"
  top: "fc3_4/sqz"
}
layer {
  name: "fc3_4/exc"
  type: "InnerProduct"
  bottom: "fc3_4/sqz"
  top: "fc3_4/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm3_4/gate"
  type: "Sigmoid"
  bottom: "fc3_4/exc"
  top: "fc3_4/exc"
}
layer {
  name: "scale3_4"
  type: "Scale"
  bottom: "conv3_4/x3"
  bottom: "fc3_4/exc"
  top: "scale3_4"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_3_4"
  type: "Eltwise"
  bottom: "block_3_3"
  bottom: "scale3_4"
  top: "block_3_4"
}
layer {
  name: "relu3_4"
  type: "ReLU"
  bottom: "block_3_4"
  top: "block_3_4"
}
layer {
  name: "conv4_1/prj"
  type: "Convolution"
  bottom: "block_3_4"
  top: "conv4_1/prj"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/prj/bn"
  type: "BatchNorm"
  bottom: "conv4_1/prj"
  top: "conv4_1/prj"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_1/prj/scale"
  type: "Scale"
  bottom: "conv4_1/prj"
  top: "conv4_1/prj"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv4_1/x1"
  type: "Convolution"
  bottom: "block_3_4"
  top: "conv4_1/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/x1/bn"
  type: "BatchNorm"
  bottom: "conv4_1/x1"
  top: "conv4_1/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_1/x1/scale"
  type: "Scale"
  bottom: "conv4_1/x1"
  top: "conv4_1/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_1/x1"
  type: "ReLU"
  bottom: "conv4_1/x1"
  top: "conv4_1/x1"
}
layer {
  name: "conv4_1/x2"
  type: "Convolution"
  bottom: "conv4_1/x1"
  top: "conv4_1/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/x2/bn"
  type: "BatchNorm"
  bottom: "conv4_1/x2"
  top: "conv4_1/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_1/x2/scale"
  type: "Scale"
  bottom: "conv4_1/x2"
  top: "conv4_1/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_1/x2"
  type: "ReLU"
  bottom: "conv4_1/x2"
  top: "conv4_1/x2"
}
layer {
  name: "conv4_1/x3"
  type: "Convolution"
  bottom: "conv4_1/x2"
  top: "conv4_1/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_1/x3/bn"
  type: "BatchNorm"
  bottom: "conv4_1/x3"
  top: "conv4_1/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_1/x3/scale"
  type: "Scale"
  bottom: "conv4_1/x3"
  top: "conv4_1/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool4_1/gap"
  type: "Pooling"
  bottom: "conv4_1/x3"
  top: "pool4_1/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc4_1/sqz"
  type: "InnerProduct"
  bottom: "pool4_1/gap"
  top: "fc4_1/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 64
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_1/sqz"
  type: "ReLU"
  bottom: "fc4_1/sqz"
  top: "fc4_1/sqz"
}
layer {
  name: "fc4_1/exc"
  type: "InnerProduct"
  bottom: "fc4_1/sqz"
  top: "fc4_1/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm4_1/gate"
  type: "Sigmoid"
  bottom: "fc4_1/exc"
  top: "fc4_1/exc"
}
layer {
  name: "scale4_1"
  type: "Scale"
  bottom: "conv4_1/x3"
  bottom: "fc4_1/exc"
  top: "scale4_1"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_4_1"
  type: "Eltwise"
  bottom: "conv4_1/prj"
  bottom: "scale4_1"
  top: "block_4_1"
}
layer {
  name: "relu4_1"
  type: "ReLU"
  bottom: "block_4_1"
  top: "block_4_1"
}
layer {
  name: "conv4_2/x1"
  type: "Convolution"
  bottom: "block_4_1"
  top: "conv4_2/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_2/x1/bn"
  type: "BatchNorm"
  bottom: "conv4_2/x1"
  top: "conv4_2/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_2/x1/scale"
  type: "Scale"
  bottom: "conv4_2/x1"
  top: "conv4_2/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_2/x1"
  type: "ReLU"
  bottom: "conv4_2/x1"
  top: "conv4_2/x1"
}
layer {
  name: "conv4_2/x2"
  type: "Convolution"
  bottom: "conv4_2/x1"
  top: "conv4_2/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_2/x2/bn"
  type: "BatchNorm"
  bottom: "conv4_2/x2"
  top: "conv4_2/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_2/x2/scale"
  type: "Scale"
  bottom: "conv4_2/x2"
  top: "conv4_2/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_2/x2"
  type: "ReLU"
  bottom: "conv4_2/x2"
  top: "conv4_2/x2"
}
layer {
  name: "conv4_2/x3"
  type: "Convolution"
  bottom: "conv4_2/x2"
  top: "conv4_2/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_2/x3/bn"
  type: "BatchNorm"
  bottom: "conv4_2/x3"
  top: "conv4_2/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_2/x3/scale"
  type: "Scale"
  bottom: "conv4_2/x3"
  top: "conv4_2/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool4_2/gap"
  type: "Pooling"
  bottom: "conv4_2/x3"
  top: "pool4_2/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc4_2/sqz"
  type: "InnerProduct"
  bottom: "pool4_2/gap"
  top: "fc4_2/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 64
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_2/sqz"
  type: "ReLU"
  bottom: "fc4_2/sqz"
  top: "fc4_2/sqz"
}
layer {
  name: "fc4_2/exc"
  type: "InnerProduct"
  bottom: "fc4_2/sqz"
  top: "fc4_2/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm4_2/gate"
  type: "Sigmoid"
  bottom: "fc4_2/exc"
  top: "fc4_2/exc"
}
layer {
  name: "scale4_2"
  type: "Scale"
  bottom: "conv4_2/x3"
  bottom: "fc4_2/exc"
  top: "scale4_2"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_4_2"
  type: "Eltwise"
  bottom: "block_4_1"
  bottom: "scale4_2"
  top: "block_4_2"
}
layer {
  name: "relu4_2"
  type: "ReLU"
  bottom: "block_4_2"
  top: "block_4_2"
}
layer {
  name: "conv4_3/x1"
  type: "Convolution"
  bottom: "block_4_2"
  top: "conv4_3/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_3/x1/bn"
  type: "BatchNorm"
  bottom: "conv4_3/x1"
  top: "conv4_3/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_3/x1/scale"
  type: "Scale"
  bottom: "conv4_3/x1"
  top: "conv4_3/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_3/x1"
  type: "ReLU"
  bottom: "conv4_3/x1"
  top: "conv4_3/x1"
}
layer {
  name: "conv4_3/x2"
  type: "Convolution"
  bottom: "conv4_3/x1"
  top: "conv4_3/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_3/x2/bn"
  type: "BatchNorm"
  bottom: "conv4_3/x2"
  top: "conv4_3/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_3/x2/scale"
  type: "Scale"
  bottom: "conv4_3/x2"
  top: "conv4_3/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_3/x2"
  type: "ReLU"
  bottom: "conv4_3/x2"
  top: "conv4_3/x2"
}
layer {
  name: "conv4_3/x3"
  type: "Convolution"
  bottom: "conv4_3/x2"
  top: "conv4_3/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_3/x3/bn"
  type: "BatchNorm"
  bottom: "conv4_3/x3"
  top: "conv4_3/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_3/x3/scale"
  type: "Scale"
  bottom: "conv4_3/x3"
  top: "conv4_3/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool4_3/gap"
  type: "Pooling"
  bottom: "conv4_3/x3"
  top: "pool4_3/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc4_3/sqz"
  type: "InnerProduct"
  bottom: "pool4_3/gap"
  top: "fc4_3/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 64
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_3/sqz"
  type: "ReLU"
  bottom: "fc4_3/sqz"
  top: "fc4_3/sqz"
}
layer {
  name: "fc4_3/exc"
  type: "InnerProduct"
  bottom: "fc4_3/sqz"
  top: "fc4_3/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm4_3/gate"
  type: "Sigmoid"
  bottom: "fc4_3/exc"
  top: "fc4_3/exc"
}
layer {
  name: "scale4_3"
  type: "Scale"
  bottom: "conv4_3/x3"
  bottom: "fc4_3/exc"
  top: "scale4_3"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_4_3"
  type: "Eltwise"
  bottom: "block_4_2"
  bottom: "scale4_3"
  top: "block_4_3"
}
layer {
  name: "relu4_3"
  type: "ReLU"
  bottom: "block_4_3"
  top: "block_4_3"
}
layer {
  name: "conv4_4/x1"
  type: "Convolution"
  bottom: "block_4_3"
  top: "conv4_4/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_4/x1/bn"
  type: "BatchNorm"
  bottom: "conv4_4/x1"
  top: "conv4_4/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_4/x1/scale"
  type: "Scale"
  bottom: "conv4_4/x1"
  top: "conv4_4/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_4/x1"
  type: "ReLU"
  bottom: "conv4_4/x1"
  top: "conv4_4/x1"
}
layer {
  name: "conv4_4/x2"
  type: "Convolution"
  bottom: "conv4_4/x1"
  top: "conv4_4/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_4/x2/bn"
  type: "BatchNorm"
  bottom: "conv4_4/x2"
  top: "conv4_4/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_4/x2/scale"
  type: "Scale"
  bottom: "conv4_4/x2"
  top: "conv4_4/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_4/x2"
  type: "ReLU"
  bottom: "conv4_4/x2"
  top: "conv4_4/x2"
}
layer {
  name: "conv4_4/x3"
  type: "Convolution"
  bottom: "conv4_4/x2"
  top: "conv4_4/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_4/x3/bn"
  type: "BatchNorm"
  bottom: "conv4_4/x3"
  top: "conv4_4/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_4/x3/scale"
  type: "Scale"
  bottom: "conv4_4/x3"
  top: "conv4_4/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool4_4/gap"
  type: "Pooling"
  bottom: "conv4_4/x3"
  top: "pool4_4/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc4_4/sqz"
  type: "InnerProduct"
  bottom: "pool4_4/gap"
  top: "fc4_4/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 64
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_4/sqz"
  type: "ReLU"
  bottom: "fc4_4/sqz"
  top: "fc4_4/sqz"
}
layer {
  name: "fc4_4/exc"
  type: "InnerProduct"
  bottom: "fc4_4/sqz"
  top: "fc4_4/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm4_4/gate"
  type: "Sigmoid"
  bottom: "fc4_4/exc"
  top: "fc4_4/exc"
}
layer {
  name: "scale4_4"
  type: "Scale"
  bottom: "conv4_4/x3"
  bottom: "fc4_4/exc"
  top: "scale4_4"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_4_4"
  type: "Eltwise"
  bottom: "block_4_3"
  bottom: "scale4_4"
  top: "block_4_4"
}
layer {
  name: "relu4_4"
  type: "ReLU"
  bottom: "block_4_4"
  top: "block_4_4"
}
layer {
  name: "conv4_5/x1"
  type: "Convolution"
  bottom: "block_4_4"
  top: "conv4_5/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_5/x1/bn"
  type: "BatchNorm"
  bottom: "conv4_5/x1"
  top: "conv4_5/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_5/x1/scale"
  type: "Scale"
  bottom: "conv4_5/x1"
  top: "conv4_5/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_5/x1"
  type: "ReLU"
  bottom: "conv4_5/x1"
  top: "conv4_5/x1"
}
layer {
  name: "conv4_5/x2"
  type: "Convolution"
  bottom: "conv4_5/x1"
  top: "conv4_5/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_5/x2/bn"
  type: "BatchNorm"
  bottom: "conv4_5/x2"
  top: "conv4_5/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_5/x2/scale"
  type: "Scale"
  bottom: "conv4_5/x2"
  top: "conv4_5/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_5/x2"
  type: "ReLU"
  bottom: "conv4_5/x2"
  top: "conv4_5/x2"
}
layer {
  name: "conv4_5/x3"
  type: "Convolution"
  bottom: "conv4_5/x2"
  top: "conv4_5/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_5/x3/bn"
  type: "BatchNorm"
  bottom: "conv4_5/x3"
  top: "conv4_5/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_5/x3/scale"
  type: "Scale"
  bottom: "conv4_5/x3"
  top: "conv4_5/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool4_5/gap"
  type: "Pooling"
  bottom: "conv4_5/x3"
  top: "pool4_5/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc4_5/sqz"
  type: "InnerProduct"
  bottom: "pool4_5/gap"
  top: "fc4_5/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 64
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_5/sqz"
  type: "ReLU"
  bottom: "fc4_5/sqz"
  top: "fc4_5/sqz"
}
layer {
  name: "fc4_5/exc"
  type: "InnerProduct"
  bottom: "fc4_5/sqz"
  top: "fc4_5/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm4_5/gate"
  type: "Sigmoid"
  bottom: "fc4_5/exc"
  top: "fc4_5/exc"
}
layer {
  name: "scale4_5"
  type: "Scale"
  bottom: "conv4_5/x3"
  bottom: "fc4_5/exc"
  top: "scale4_5"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_4_5"
  type: "Eltwise"
  bottom: "block_4_4"
  bottom: "scale4_5"
  top: "block_4_5"
}
layer {
  name: "relu4_5"
  type: "ReLU"
  bottom: "block_4_5"
  top: "block_4_5"
}
layer {
  name: "conv4_6/x1"
  type: "Convolution"
  bottom: "block_4_5"
  top: "conv4_6/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_6/x1/bn"
  type: "BatchNorm"
  bottom: "conv4_6/x1"
  top: "conv4_6/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_6/x1/scale"
  type: "Scale"
  bottom: "conv4_6/x1"
  top: "conv4_6/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_6/x1"
  type: "ReLU"
  bottom: "conv4_6/x1"
  top: "conv4_6/x1"
}
layer {
  name: "conv4_6/x2"
  type: "Convolution"
  bottom: "conv4_6/x1"
  top: "conv4_6/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 256
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_6/x2/bn"
  type: "BatchNorm"
  bottom: "conv4_6/x2"
  top: "conv4_6/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_6/x2/scale"
  type: "Scale"
  bottom: "conv4_6/x2"
  top: "conv4_6/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu4_6/x2"
  type: "ReLU"
  bottom: "conv4_6/x2"
  top: "conv4_6/x2"
}
layer {
  name: "conv4_6/x3"
  type: "Convolution"
  bottom: "conv4_6/x2"
  top: "conv4_6/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 1024
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv4_6/x3/bn"
  type: "BatchNorm"
  bottom: "conv4_6/x3"
  top: "conv4_6/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv4_6/x3/scale"
  type: "Scale"
  bottom: "conv4_6/x3"
  top: "conv4_6/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool4_6/gap"
  type: "Pooling"
  bottom: "conv4_6/x3"
  top: "pool4_6/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc4_6/sqz"
  type: "InnerProduct"
  bottom: "pool4_6/gap"
  top: "fc4_6/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 64
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_6/sqz"
  type: "ReLU"
  bottom: "fc4_6/sqz"
  top: "fc4_6/sqz"
}
layer {
  name: "fc4_6/exc"
  type: "InnerProduct"
  bottom: "fc4_6/sqz"
  top: "fc4_6/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm4_6/gate"
  type: "Sigmoid"
  bottom: "fc4_6/exc"
  top: "fc4_6/exc"
}
layer {
  name: "scale4_6"
  type: "Scale"
  bottom: "conv4_6/x3"
  bottom: "fc4_6/exc"
  top: "scale4_6"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_4_6"
  type: "Eltwise"
  bottom: "block_4_5"
  bottom: "scale4_6"
  top: "block_4_6"
}
layer {
  name: "relu4_6"
  type: "ReLU"
  bottom: "block_4_6"
  top: "block_4_6"
}
layer {
  name: "conv5_1/prj"
  type: "Convolution"
  bottom: "block_4_6"
  top: "conv5_1/prj"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 2048
    bias_term: false
    kernel_size: 1
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/prj/bn"
  type: "BatchNorm"
  bottom: "conv5_1/prj"
  top: "conv5_1/prj"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_1/prj/scale"
  type: "Scale"
  bottom: "conv5_1/prj"
  top: "conv5_1/prj"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "conv5_1/x1"
  type: "Convolution"
  bottom: "block_4_6"
  top: "conv5_1/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/x1/bn"
  type: "BatchNorm"
  bottom: "conv5_1/x1"
  top: "conv5_1/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_1/x1/scale"
  type: "Scale"
  bottom: "conv5_1/x1"
  top: "conv5_1/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5_1/x1"
  type: "ReLU"
  bottom: "conv5_1/x1"
  top: "conv5_1/x1"
}
layer {
  name: "conv5_1/x2"
  type: "Convolution"
  bottom: "conv5_1/x1"
  top: "conv5_1/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    stride: 2
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/x2/bn"
  type: "BatchNorm"
  bottom: "conv5_1/x2"
  top: "conv5_1/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_1/x2/scale"
  type: "Scale"
  bottom: "conv5_1/x2"
  top: "conv5_1/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5_1/x2"
  type: "ReLU"
  bottom: "conv5_1/x2"
  top: "conv5_1/x2"
}
layer {
  name: "conv5_1/x3"
  type: "Convolution"
  bottom: "conv5_1/x2"
  top: "conv5_1/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 2048
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_1/x3/bn"
  type: "BatchNorm"
  bottom: "conv5_1/x3"
  top: "conv5_1/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_1/x3/scale"
  type: "Scale"
  bottom: "conv5_1/x3"
  top: "conv5_1/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool5_1/gap"
  type: "Pooling"
  bottom: "conv5_1/x3"
  top: "pool5_1/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc5_1/sqz"
  type: "InnerProduct"
  bottom: "pool5_1/gap"
  top: "fc5_1/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 128
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu5_1/sqz"
  type: "ReLU"
  bottom: "fc5_1/sqz"
  top: "fc5_1/sqz"
}
layer {
  name: "fc5_1/exc"
  type: "InnerProduct"
  bottom: "fc5_1/sqz"
  top: "fc5_1/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 2048
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm5_1/gate"
  type: "Sigmoid"
  bottom: "fc5_1/exc"
  top: "fc5_1/exc"
}
layer {
  name: "scale5_1"
  type: "Scale"
  bottom: "conv5_1/x3"
  bottom: "fc5_1/exc"
  top: "scale5_1"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_5_1"
  type: "Eltwise"
  bottom: "conv5_1/prj"
  bottom: "scale5_1"
  top: "block_5_1"
}
layer {
  name: "relu5_1"
  type: "ReLU"
  bottom: "block_5_1"
  top: "block_5_1"
}
layer {
  name: "conv5_2/x1"
  type: "Convolution"
  bottom: "block_5_1"
  top: "conv5_2/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_2/x1/bn"
  type: "BatchNorm"
  bottom: "conv5_2/x1"
  top: "conv5_2/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_2/x1/scale"
  type: "Scale"
  bottom: "conv5_2/x1"
  top: "conv5_2/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5_2/x1"
  type: "ReLU"
  bottom: "conv5_2/x1"
  top: "conv5_2/x1"
}
layer {
  name: "conv5_2/x2"
  type: "Convolution"
  bottom: "conv5_2/x1"
  top: "conv5_2/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_2/x2/bn"
  type: "BatchNorm"
  bottom: "conv5_2/x2"
  top: "conv5_2/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_2/x2/scale"
  type: "Scale"
  bottom: "conv5_2/x2"
  top: "conv5_2/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5_2/x2"
  type: "ReLU"
  bottom: "conv5_2/x2"
  top: "conv5_2/x2"
}
layer {
  name: "conv5_2/x3"
  type: "Convolution"
  bottom: "conv5_2/x2"
  top: "conv5_2/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 2048
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_2/x3/bn"
  type: "BatchNorm"
  bottom: "conv5_2/x3"
  top: "conv5_2/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_2/x3/scale"
  type: "Scale"
  bottom: "conv5_2/x3"
  top: "conv5_2/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool5_2/gap"
  type: "Pooling"
  bottom: "conv5_2/x3"
  top: "pool5_2/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc5_2/sqz"
  type: "InnerProduct"
  bottom: "pool5_2/gap"
  top: "fc5_2/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 128
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu5_2/sqz"
  type: "ReLU"
  bottom: "fc5_2/sqz"
  top: "fc5_2/sqz"
}
layer {
  name: "fc5_2/exc"
  type: "InnerProduct"
  bottom: "fc5_2/sqz"
  top: "fc5_2/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 2048
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm5_2/gate"
  type: "Sigmoid"
  bottom: "fc5_2/exc"
  top: "fc5_2/exc"
}
layer {
  name: "scale5_2"
  type: "Scale"
  bottom: "conv5_2/x3"
  bottom: "fc5_2/exc"
  top: "scale5_2"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_5_2"
  type: "Eltwise"
  bottom: "block_5_1"
  bottom: "scale5_2"
  top: "block_5_2"
}
layer {
  name: "relu5_2"
  type: "ReLU"
  bottom: "block_5_2"
  top: "block_5_2"
}
layer {
  name: "conv5_3/x1"
  type: "Convolution"
  bottom: "block_5_2"
  top: "conv5_3/x1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_3/x1/bn"
  type: "BatchNorm"
  bottom: "conv5_3/x1"
  top: "conv5_3/x1"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_3/x1/scale"
  type: "Scale"
  bottom: "conv5_3/x1"
  top: "conv5_3/x1"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5_3/x1"
  type: "ReLU"
  bottom: "conv5_3/x1"
  top: "conv5_3/x1"
}
layer {
  name: "conv5_3/x2"
  type: "Convolution"
  bottom: "conv5_3/x1"
  top: "conv5_3/x2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 512
    bias_term: false
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_3/x2/bn"
  type: "BatchNorm"
  bottom: "conv5_3/x2"
  top: "conv5_3/x2"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_3/x2/scale"
  type: "Scale"
  bottom: "conv5_3/x2"
  top: "conv5_3/x2"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "relu5_3/x2"
  type: "ReLU"
  bottom: "conv5_3/x2"
  top: "conv5_3/x2"
}
layer {
  name: "conv5_3/x3"
  type: "Convolution"
  bottom: "conv5_3/x2"
  top: "conv5_3/x3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 2048
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv5_3/x3/bn"
  type: "BatchNorm"
  bottom: "conv5_3/x3"
  top: "conv5_3/x3"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  param {
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    eps: 1e-4
  }
}
layer {
  name: "conv5_3/x3/scale"
  type: "Scale"
  bottom: "conv5_3/x3"
  top: "conv5_3/x3"
  scale_param {
    bias_term: true
  }
}
layer {
  name: "pool5_3/gap"
  type: "Pooling"
  bottom: "conv5_3/x3"
  top: "pool5_3/gap"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc5_3/sqz"
  type: "InnerProduct"
  bottom: "pool5_3/gap"
  top: "fc5_3/sqz"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 128
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu5_3/sqz"
  type: "ReLU"
  bottom: "fc5_3/sqz"
  top: "fc5_3/sqz"
}
layer {
  name: "fc5_3/exc"
  type: "InnerProduct"
  bottom: "fc5_3/sqz"
  top: "fc5_3/exc"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 2048
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "sigm5_3/gate"
  type: "Sigmoid"
  bottom: "fc5_3/exc"
  top: "fc5_3/exc"
}
layer {
  name: "scale5_3"
  type: "Scale"
  bottom: "conv5_3/x3"
  bottom: "fc5_3/exc"
  top: "scale5_3"
  scale_param {
    axis: 0
    bias_term: false
  }
}
layer {
  name: "block_5_3"
  type: "Eltwise"
  bottom: "block_5_2"
  bottom: "scale5_3"
  top: "block_5_3"
}
layer {
  name: "relu5_3"
  type: "ReLU"
  bottom: "block_5_3"
  top: "block_5_3"
}
layer {
  name: "pool5"
  type: "Pooling"
  bottom: "block_5_3"
  top: "pool5"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc6"
  type: "Convolution"
  bottom: "pool5"
  top: "fc6"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  convolution_param {
    num_output: 1000
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "prob"
  top: "prob"
  type: "Softmax"
  bottom: "fc6"
}
#layer {
#  name: "loss"
#  type: "SoftmaxWithLoss"
#  bottom: "fc6"
#  bottom: "label"
#  top: "loss"
#}
#layer {
#  name: "top1/acc"
#  type: "Accuracy"
#  bottom: "fc6"
#  bottom: "label"
#  top: "top1/acc"
#  include {
#    phase: TEST
#  }
#}
#layer {
#  name: "top5/acc"
#  type: "Accuracy"
#  bottom: "fc6"
#  bottom: "label"
#  top: "top5/acc"
#  include {
#    phase: TEST
#  }
#  accuracy_param {
#    top_k: 5
#  }
#}