name: "ResNet-18"

layer {
  name: "data"
  type: "Data"
  top: "data"
  top: "label"
  transform_param {
    crop_size: 224
    mean_value: 128
    mean_value: 128
    mean_value: 128
    mirror: true
  }
  data_param {
    source: "examples/imagenet/ilsvrc12_train_lmdb"
    batch_size: 128
    backend: LMDB
  }
  include {
    phase: TRAIN
  }
}
  
layer {
  name: "data"
  type: "Data"
  top: "data"
  top: "label"
  transform_param {
    crop_size: 224
    mean_value: 128
    mean_value: 128
    mean_value: 128
    mirror: false
  }
  data_param {
   source: "examples/imagenet/ilsvrc12_val_lmdb"
   batch_size: 50
   backend: LMDB
  }
  include {
    phase: TEST
  }
}

layer {
  name: "conv1"
  bottom: "data"
  top: "conv1"
  type: "Convolution"
  param {
   lr_mult: 1 
   decay_mult: 1 
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 7
    pad: 3
    stride: 2
    weight_filler {
      type: "msra" 
      std: 0.010
    }
    bias_filler { 
      type: "constant" 
      value: 0 
    }
  }
}
layer {
  name: "bn_conv1"
  bottom: "conv1"
  top: "conv1/bn"
  type: "BatchNorm"
  param { #scale
    lr_mult: 1
    decay_mult: 1
  }
  param { #shift/bias
    lr_mult: 1
    decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
       type: "constant"
       value: 1
    }
    bias_filler {
       type: "constant"
       value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "conv1_relu"
  bottom: "conv1/bn"
  top: "conv1/bn"
  type: "ReLU"
}
layer {
  name: "pool1"
  bottom: "conv1/bn"
  top: "pool1"
  type: "Pooling"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "res2a_branch2a"
  bottom: "pool1"
  top: "res2a_branch2a"
  type: "Convolution"
  param { 
   lr_mult: 1
   decay_mult: 1 
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn2a_branch2a"
  bottom: "res2a_branch2a"
  top: "res2a_branch2a/bn"
  type: "BatchNorm"
  param { #scale
    lr_mult: 1
    decay_mult: 1
  }
  param { #shift/bias
    lr_mult: 1
    decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
      decay_mult: 0
    }
  param { #global var
    lr_mult: 0
      decay_mult: 0
    }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res2a_branch2a_relu"
  bottom: "res2a_branch2a/bn"
  top: "res2a_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res2a_branch2b"
  bottom: "res2a_branch2a/bn"
  top: "res2a_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn2a_branch2b"
  bottom: "res2a_branch2b"
  top: "res2a_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res2a"
  bottom: "pool1"
  bottom: "res2a_branch2b/bn"
  top: "res2a"
  type: "Eltwise"
}
layer {
  name: "res2a_relu"
  bottom: "res2a"
  top: "res2a"
  type: "ReLU"
}
layer {
  name: "res2b_branch2a"
  bottom: "res2a"
  top: "res2b_branch2a"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn2b_branch2a"
  bottom: "res2b_branch2a"
  top: "res2b_branch2a/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res2b_branch2a_relu"
  bottom: "res2b_branch2a/bn"
  top: "res2b_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res2b_branch2b"
  bottom: "res2b_branch2a/bn"
  top: "res2b_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:64
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn2b_branch2b"
  bottom: "res2b_branch2b"
  top: "res2b_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res2b"
  bottom: "res2a"
  bottom: "res2b_branch2b/bn"
  top: "res2b"
  type: "Eltwise"
}
layer {
  name: "res2b_relu"
  bottom: "res2b"
  top: "res2b"
  type: "ReLU"
}
layer {
  name: "res3a_branch2a"
  bottom: "res2b"
  top: "res3a_branch2a"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:128
    kernel_size: 3
    pad: 1
    stride: 2
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn3a_branch2a"
  bottom: "res3a_branch2a"
  top: "res3a_branch2a/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res3a_branch2a_relu"
  bottom: "res3a_branch2a/bn"
  top: "res3a_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res3a_branch2b"
  bottom: "res3a_branch2a/bn"
  top: "res3a_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:128
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn3a_branch2b"
  bottom: "res3a_branch2b"
  top: "res3a_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res3a_branch1"
  bottom: "res2b"
  top: "res3a_branch1"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:128
    kernel_size: 1
    pad: 0
    stride: 2
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn3a_branch1"
  bottom: "res3a_branch1"
  top: "res3a_branch1/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res3a"
  bottom: "res3a_branch1/bn"
  bottom: "res3a_branch2b/bn"
  top: "res3a"
  type: "Eltwise"
}
layer {
  name: "res3a_relu"
  bottom: "res3a"
  top: "res3a"
  type: "ReLU"
}
layer {
  name: "res3b_branch2a"
  bottom: "res3a"
  top: "res3b_branch2a"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:128
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn3b_branch2a"
  bottom: "res3b_branch2a"
  top: "res3b_branch2a/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res3b_branch2a_relu"
  bottom: "res3b_branch2a/bn"
  top: "res3b_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res3b_branch2b"
  bottom: "res3b_branch2a/bn"
  top: "res3b_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:128
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn3b_branch2b"
  bottom: "res3b_branch2b"
  top: "res3b_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res3b"
  bottom: "res3a"
  bottom: "res3b_branch2b/bn"
  top: "res3b"
  type: "Eltwise"
}
layer {
  name: "res3b_relu"
  bottom: "res3b"
  top: "res3b"
  type: "ReLU"
}
layer {
  name: "res4a_branch2a"
  bottom: "res3b"
  top: "res4a_branch2a"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:256
    kernel_size: 3
    pad: 1
    stride: 2
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn4a_branch2a"
  bottom: "res4a_branch2a"
  top: "res4a_branch2a/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res4a_branch2a_relu"
  bottom: "res4a_branch2a/bn"
  top: "res4a_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res4a_branch2b"
  bottom: "res4a_branch2a/bn"
  top: "res4a_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:256
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn4a_branch2b"
  bottom: "res4a_branch2b"
  top: "res4a_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res4a_branch1"
  bottom: "res3b"
  top: "res4a_branch1"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:256
    kernel_size: 1
    pad: 0
    stride: 2
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn4a_branch1"
  bottom: "res4a_branch1"
  top: "res4a_branch1/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res4a"
  bottom: "res4a_branch1/bn"
  bottom: "res4a_branch2b/bn"
  top: "res4a"
  type: "Eltwise"
}
layer {
  name: "res4a_relu"
  bottom: "res4a"
  top: "res4a"
  type: "ReLU"
}
layer {
  name: "res4b_branch2a"
  bottom: "res4a"
  top: "res4b_branch2a"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:256
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn4b_branch2a"
  bottom: "res4b_branch2a"
  top: "res4b_branch2a/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res4b_branch2a_relu"
  bottom: "res4b_branch2a/bn"
  top: "res4b_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res4b_branch2b"
  bottom: "res4b_branch2a/bn"
  top: "res4b_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn4b_branch2b"
  bottom: "res4b_branch2b"
  top: "res4b_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res4b"
  bottom: "res4a"
  bottom: "res4b_branch2b/bn"
  top: "res4b"
  type: "Eltwise"
}
layer {
  name: "res4b_relu"
  bottom: "res4b"
  top: "res4b"
  type: "ReLU"
}
layer {
  name: "res5a_branch2a"
  bottom: "res4b"
  top: "res5a_branch2a"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    pad: 1
    stride: 2
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn5a_branch2a"
  bottom: "res5a_branch2a"
  top: "res5a_branch2a/bn"
  type: "BatchNorm"
  param { #scale
    lr_mult: 1
    decay_mult: 1
  }
  param { #shift/bias
    lr_mult: 1
    decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res5a_branch2a_relu"
  bottom: "res5a_branch2a/bn"
  top: "res5a_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res5a_branch2b"
  bottom: "res5a_branch2a/bn"
  top: "res5a_branch2b"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output:512
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn5a_branch2b"
  bottom: "res5a_branch2b"
  top: "res5a_branch2b/bn"
  type: "BatchNorm"
  param { #scale
  lr_mult: 1
  decay_mult: 1
  }
  param { #shift/bias
  lr_mult: 1
  decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res5a_branch1"
  bottom: "res4b"
  top: "res5a_branch1"
  type: "Convolution"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  convolution_param {
    num_output: 512
    kernel_size: 1
    pad: 0
    stride: 2
    bias_term: false
    weight_filler {
       type: "msra" 
       std: 0.010 
    }
    dilation: 1
    group: 1
   }
}
layer {
  name: "bn5a_branch1"
  bottom: "res5a_branch1"
  top: "res5a_branch1/bn"
  type: "BatchNorm"
  param { #scale
    lr_mult: 1
    decay_mult: 1
  }
  param { #shift/bias
    lr_mult: 1
    decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res5a"
  bottom: "res5a_branch1/bn"
  bottom: "res5a_branch2b/bn"
  top: "res5a"
  type: "Eltwise"
}
layer {
  name: "res5a_relu"
  bottom: "res5a"
  top: "res5a"
  type: "ReLU"
}
layer {
  name: "res5b_branch2a"
  bottom: "res5a"
  top: "res5b_branch2a"
  type: "Convolution"
  param {
   lr_mult: 1 
   decay_mult: 1 
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra"
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn5b_branch2a"
  bottom: "res5b_branch2a"
  top: "res5b_branch2a/bn"
  type: "BatchNorm"
  param { #scale
    lr_mult: 1
    decay_mult: 1
  }
  param { #shift/bias
    lr_mult: 1
    decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res5b_branch2a_relu"
  bottom: "res5b_branch2a/bn"
  top: "res5b_branch2a/bn"
  type: "ReLU"
}
layer {
  name: "res5b_branch2b"
  bottom: "res5b_branch2a/bn"
  top: "res5b_branch2b"
  type: "Convolution"
  param {
   lr_mult: 1 
   decay_mult: 1 
  }
  convolution_param {
    num_output: 512
    kernel_size: 3
    pad: 1
    stride: 1
    bias_term: false
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    dilation: 1
    group: 1
  }
}
layer {
  name: "bn5b_branch2b"
  bottom: "res5b_branch2b"
  top: "res5b_branch2b/bn"
  type: "BatchNorm"
  param {  #scale
    lr_mult: 1
    decay_mult: 1
  }
  param { #shift/bias
    lr_mult: 1
    decay_mult: 1
  } 
  param { #global mean
    lr_mult: 0
    decay_mult: 0
  }
  param { #global var
    lr_mult: 0
    decay_mult: 0
  }
  batch_norm_param {
    moving_average_fraction: 0.9
    eps: 0.0001
    scale_filler {
      type: "constant"
      value: 1
    }
    bias_filler {
      type: "constant"
      value: 0
    }
    engine: CUDNN
  }
}
layer {
  name: "res5b"
  bottom: "res5a"
  bottom: "res5b_branch2b/bn"
  top: "res5b"
  type: "Eltwise"
}
layer {
  name: "res5b_relu"
  bottom: "res5b"
  top: "res5b"
  type: "ReLU"
}
layer {
  name: "pool5"
  bottom: "res5b"
  top: "pool5"
  type: "Pooling"
  pooling_param {
    pool: AVE
    kernel_size: 7
    stride: 1
  }
}

#layer {
#  name: "conv6"
#  bottom: "pool5"
#  top: "conv6"
#  type: "Convolution"
#  param { 
#    lr_mult: 1
#    decay_mult: 1 
#  }
#  param {
#     lr_mult: 2 
#    decay_mult: 0
#  }
#  convolution_param {
#    num_output: 1000
#    kernel_size: 1
#    pad: 0
#    stride: 1
#    weight_filler {
#      type: "msra" 
#      std: 0.010 
#    }
#    bias_filler { 
#      type: "constant"
#      value: 0 
#    }
#  }
#}

layer {
  type: "InnerProduct"
  name: "ip"
  bottom: "pool5"
  top: "ip"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1000
    weight_filler {
      type: "msra" 
      std: 0.010 
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}


layer {
  name: "loss"
  type: "SoftmaxWithLoss"
#  bottom: "conv6"
  bottom: "ip"
  bottom: "label"
  propagate_down: 1
  propagate_down: 0
  top: "loss"
  loss_weight: 1
}
layer {
  name: "accuracy"
  type: "Accuracy"
#  bottom: "conv6"
  bottom: "ip"
  bottom: "label"
  top: "accuracy/top-1"
  include {
    phase: TEST
  }
}
layer {
  name: "accuracy/top-5"
  type: "Accuracy"
#  bottom: "conv6"
  bottom: "ip"
  bottom: "label"
  top: "accuracy/top-5"
  include {
    phase: TEST
  }
  accuracy_param {
    top_k: 5
  }
}