train_val_cfn_jps++.prototxt

name: "CaffeNet"
layer {
  name: "data_0"
  type: "Data"
  top: "data_0"
  top: "label"
  include {
    phase: TRAIN
  }
  transform_param {
    mirror: false
    crop_size: 255
    hash_file: "./permutations_701.bin"
    siamese_size: 9
    patch_size: 64
    jitter_size: 18
    #gap_size: 16
   
  }
  data_param {
    source: "./jps_dataset/train"
    batch_size: 256
    backend: LMDB
    rand_jump: 2
  }
}
layer {
  name: "data_0"
  type: "Data"
  top: "data_0"
  top: "label"
  include {
    phase: TEST
  }
  transform_param {
    mirror: false
    crop_size: 255
    hash_file: "./permutations_701.bin"
    siamese_size: 9
    patch_size: 64
    jitter_size: 18
  }
  data_param {
    source: "./jps_dataset/val"
    batch_size: 50
    backend: LMDB
    rand_jump: 0
  }
}

layer {
  name: "data_r"
  type: "Data"
  top: "data_r"
  #top: "label"
  include {
    phase: TRAIN
  }
  transform_param {
    mirror: false
    crop_size: 255
    siamese_size: 9
    patch_size: 64
    jitter_size: 18
  }
  data_param {
    source: "./jps_dataset/train"
    batch_size: 256
    backend: LMDB
    rand_jump: 8
  }
}

layer {
  name: "data_r"
  type: "Data"
  top: "data_r"
  #top: "label"
  include {
    phase: TEST
  }
  transform_param {
    mirror: false
    crop_size: 255
    siamese_size: 9
    patch_size: 64
    jitter_size: 18
  }
  data_param {
    source: "./jps_dataset/val"
    batch_size: 50
    backend: LMDB
    rand_jump: 3
  }
}

############# Slicing layers ###########################
layer {
  name: "slice_data"
  type: "Slice"
  bottom: "data_0"
  top: "data_1_"
  top: "data_2_"
  top: "data_3_"
  top: "data_4_"
  top: "data_5_"
  top: "data_6_"
  top: "data_7_"
  top: "data_8_"
  top: "data_9_"  
  slice_param {
    slice_dim: 1
    slice_point: 3
    slice_point: 6
    slice_point: 9
    slice_point: 12
    slice_point: 15
    slice_point: 18   
    slice_point: 21
    slice_point: 24
  }
}

layer {
  name: "slice_data"
  type: "Slice"
  bottom: "data_r"
  top: "data_r1"
  top: "data_r2"
  top: "data_r3"
  top: "data_r4"
  top: "data_r5"
  top: "data_r6"
  top: "data_r7"
  top: "data_r8"
  top: "data_r9"
  slice_param {
    slice_dim: 1
    slice_point: 3
    slice_point: 6
    slice_point: 9
    slice_point: 12
    slice_point: 15
    slice_point: 18
    slice_point: 21
    slice_point: 24
  } 
} 

layer {
 name: "merge"
 type: "MergeData"

 bottom: "data_1_"
 bottom: "data_2_"
 bottom: "data_3_"
 bottom: "data_4_"
bottom: "data_5_"
bottom: "data_6_"
bottom: "data_7_"
bottom: "data_8_"
bottom: "data_9_"

bottom: "data_r1"
bottom: "data_r2"
bottom: "data_r3"
bottom: "data_r4"
bottom: "data_r5"
bottom: "data_r6"
bottom: "data_r7"
bottom: "data_r8"
bottom: "data_r9"

 top: "data_1"
 top: "data_2"
 top: "data_3"
 top: "data_4"
 top: "data_5"
 top: "data_6"
 top: "data_7"
 top: "data_8"
 top: "data_9"
 top: "sub_num"

 merge_data_param {
   min_substitute: 0
   max_substitute: 3
 }

}
layer {
 name: "silence_tile_sub_ind"
 type: "Silence"
 bottom: "sub_num"
}
############### MVN ########################################
layer {
  name: "datamvn_t1"
  bottom: "data_1"
  top: "data_1"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t1"
  type: "Power"
  bottom: "data_1"
  top: "data_1"
  power_param{
    scale: 50
  }
}

layer {
  name: "datamvn_t2"
  bottom: "data_2"
  top: "data_2"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t2"
  type: "Power"
  bottom: "data_2"
  top: "data_2"
  power_param{
    scale: 50
  }
}


layer {
  name: "datamvn_t3"
  bottom: "data_3"
  top: "data_3"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t3"
  type: "Power"
  bottom: "data_3"
  top: "data_3"
  power_param{
    scale: 50
  }
}

layer {
  name: "datamvn_t4"
  bottom: "data_4"
  top: "data_4"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t4"
  type: "Power"
  bottom: "data_4"
  top: "data_4"
  power_param{
    scale: 50
  }
}

layer {
  name: "datamvn_t5"
  bottom: "data_5"
  top: "data_5"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t5"
  type: "Power"
  bottom: "data_5"
  top: "data_5"
  power_param{
    scale: 50
  }
}

layer {
  name: "datamvn_t6"
  bottom: "data_6"
  top: "data_6"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t6"
  type: "Power"
  bottom: "data_6"
  top: "data_6"
  power_param{
    scale: 50
  }
}


layer {
  name: "datamvn_t7"
  bottom: "data_7"
  top: "data_7"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t7"
  type: "Power"
  bottom: "data_7"
  top: "data_7"
  power_param{
    scale: 50
  }
}

layer {
  name: "datamvn_t8"
  bottom: "data_8"
  top: "data_8"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t8"
  type: "Power"
  bottom: "data_8"
  top: "data_8"
  power_param{
    scale: 50
  }
}

layer {
  name: "datamvn_t9"
  bottom: "data_9"
  top: "data_9"
  type: "MVN"
  mvn_param{
    across_channels: true
  }
}
layer {
  name: "datascale_t9"
  type: "Power"
  bottom: "data_9"
  top: "data_9"
  power_param{
    scale: 50
  }
}
####### Column_1  ############################################
layer {
  name: "conv1_s1"
  type: "Convolution"
  bottom: "data_1"
  top: "conv1_s1"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s1"
  type: "ReLU"
  bottom: "conv1_s1"
  top: "conv1_s1"
}
layer {
  name: "pool1_s1"
  type: "Pooling"
  bottom: "conv1_s1"
  top: "pool1_s1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s1"
  type: "LRN"
  bottom: "pool1_s1"
  top: "norm1_s1"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s1"
  type: "Convolution"
  bottom: "norm1_s1"
  top: "conv2_s1"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s1"
  type: "ReLU"
  bottom: "conv2_s1"
  top: "conv2_s1"
}
layer {
  name: "pool2_s1"
  type: "Pooling"
  bottom: "conv2_s1"
  top: "pool2_s1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s1"
  type: "LRN"
  bottom: "pool2_s1"
  top: "norm2_s1"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s1"
  type: "Convolution"
  bottom: "norm2_s1"
  top: "conv3_s1"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s1"
  type: "ReLU"
  bottom: "conv3_s1"
  top: "conv3_s1"
}
layer {
  name: "conv4_s1"
  type: "Convolution"
  bottom: "conv3_s1"
  top: "conv4_s1"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s1"
  type: "ReLU"
  bottom: "conv4_s1"
  top: "conv4_s1"
}
layer {
  name: "conv5_s1"
  type: "Convolution"
  bottom: "conv4_s1"
  top: "conv5_s1"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s1"
  type: "ReLU"
  bottom: "conv5_s1"
  top: "conv5_s1"
}
layer {
  name: "pool5_s1"
  type: "Pooling"
  bottom: "conv5_s1"
  top: "pool5_s1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s1"
  type: "InnerProduct"
  bottom: "pool5_s1"
  top: "fc6_s1"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s1"
  type: "ReLU"
  bottom: "fc6_s1"
  top: "fc6_s1"
}
layer {
  name: "drop6_s1"
  type: "Dropout"
  bottom: "fc6_s1"
  top: "fc6_s1"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_2  ############################################
layer {
  name: "conv1_s2"
  type: "Convolution"
  bottom: "data_2"
  top: "conv1_s2"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s2"
  type: "ReLU"
  bottom: "conv1_s2"
  top: "conv1_s2"
}
layer {
  name: "pool1_s2"
  type: "Pooling"
  bottom: "conv1_s2"
  top: "pool1_s2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s2"
  type: "LRN"
  bottom: "pool1_s2"
  top: "norm1_s2"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s2"
  type: "Convolution"
  bottom: "norm1_s2"
  top: "conv2_s2"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s2"
  type: "ReLU"
  bottom: "conv2_s2"
  top: "conv2_s2"
}
layer {
  name: "pool2_s2"
  type: "Pooling"
  bottom: "conv2_s2"
  top: "pool2_s2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s2"
  type: "LRN"
  bottom: "pool2_s2"
  top: "norm2_s2"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s2"
  type: "Convolution"
  bottom: "norm2_s2"
  top: "conv3_s2"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s2"
  type: "ReLU"
  bottom: "conv3_s2"
  top: "conv3_s2"
}
layer {
  name: "conv4_s2"
  type: "Convolution"
  bottom: "conv3_s2"
  top: "conv4_s2"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s2"
  type: "ReLU"
  bottom: "conv4_s2"
  top: "conv4_s2"
}
layer {
  name: "conv5_s2"
  type: "Convolution"
  bottom: "conv4_s2"
  top: "conv5_s2"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s2"
  type: "ReLU"
  bottom: "conv5_s2"
  top: "conv5_s2"
}
layer {
  name: "pool5_s2"
  type: "Pooling"
  bottom: "conv5_s2"
  top: "pool5_s2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s2"
  type: "InnerProduct"
  bottom: "pool5_s2"
  top: "fc6_s2"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s2"
  type: "ReLU"
  bottom: "fc6_s2"
  top: "fc6_s2"
}
layer {
  name: "drop6_s2"
  type: "Dropout"
  bottom: "fc6_s2"
  top: "fc6_s2"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_3  ############################################
layer {
  name: "conv1_s3"
  type: "Convolution"
  bottom: "data_3"
  top: "conv1_s3"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s3"
  type: "ReLU"
  bottom: "conv1_s3"
  top: "conv1_s3"
}
layer {
  name: "pool1_s3"
  type: "Pooling"
  bottom: "conv1_s3"
  top: "pool1_s3"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s3"
  type: "LRN"
  bottom: "pool1_s3"
  top: "norm1_s3"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s3"
  type: "Convolution"
  bottom: "norm1_s3"
  top: "conv2_s3"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s3"
  type: "ReLU"
  bottom: "conv2_s3"
  top: "conv2_s3"
}
layer {
  name: "pool2_s3"
  type: "Pooling"
  bottom: "conv2_s3"
  top: "pool2_s3"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s3"
  type: "LRN"
  bottom: "pool2_s3"
  top: "norm2_s3"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s3"
  type: "Convolution"
  bottom: "norm2_s3"
  top: "conv3_s3"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s3"
  type: "ReLU"
  bottom: "conv3_s3"
  top: "conv3_s3"
}
layer {
  name: "conv4_s3"
  type: "Convolution"
  bottom: "conv3_s3"
  top: "conv4_s3"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s3"
  type: "ReLU"
  bottom: "conv4_s3"
  top: "conv4_s3"
}
layer {
  name: "conv5_s3"
  type: "Convolution"
  bottom: "conv4_s3"
  top: "conv5_s3"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s3"
  type: "ReLU"
  bottom: "conv5_s3"
  top: "conv5_s3"
}
layer {
  name: "pool5_s3"
  type: "Pooling"
  bottom: "conv5_s3"
  top: "pool5_s3"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s3"
  type: "InnerProduct"
  bottom: "pool5_s3"
  top: "fc6_s3"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s3"
  type: "ReLU"
  bottom: "fc6_s3"
  top: "fc6_s3"
}
layer {
  name: "drop6_s3"
  type: "Dropout"
  bottom: "fc6_s3"
  top: "fc6_s3"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_4  ############################################
layer {
  name: "conv1_s4"
  type: "Convolution"
  bottom: "data_4"
  top: "conv1_s4"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s4"
  type: "ReLU"
  bottom: "conv1_s4"
  top: "conv1_s4"
}
layer {
  name: "pool1_s4"
  type: "Pooling"
  bottom: "conv1_s4"
  top: "pool1_s4"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s4"
  type: "LRN"
  bottom: "pool1_s4"
  top: "norm1_s4"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s4"
  type: "Convolution"
  bottom: "norm1_s4"
  top: "conv2_s4"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s4"
  type: "ReLU"
  bottom: "conv2_s4"
  top: "conv2_s4"
}
layer {
  name: "pool2_s4"
  type: "Pooling"
  bottom: "conv2_s4"
  top: "pool2_s4"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s4"
  type: "LRN"
  bottom: "pool2_s4"
  top: "norm2_s4"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s4"
  type: "Convolution"
  bottom: "norm2_s4"
  top: "conv3_s4"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s4"
  type: "ReLU"
  bottom: "conv3_s4"
  top: "conv3_s4"
}
layer {
  name: "conv4_s4"
  type: "Convolution"
  bottom: "conv3_s4"
  top: "conv4_s4"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s4"
  type: "ReLU"
  bottom: "conv4_s4"
  top: "conv4_s4"
}
layer {
  name: "conv5_s4"
  type: "Convolution"
  bottom: "conv4_s4"
  top: "conv5_s4"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s4"
  type: "ReLU"
  bottom: "conv5_s4"
  top: "conv5_s4"
}
layer {
  name: "pool5_s4"
  type: "Pooling"
  bottom: "conv5_s4"
  top: "pool5_s4"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s4"
  type: "InnerProduct"
  bottom: "pool5_s4"
  top: "fc6_s4"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s4"
  type: "ReLU"
  bottom: "fc6_s4"
  top: "fc6_s4"
}
layer {
  name: "drop6_s4"
  type: "Dropout"
  bottom: "fc6_s4"
  top: "fc6_s4"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_5  ############################################
layer {
  name: "conv1_s5"
  type: "Convolution"
  bottom: "data_5"
  top: "conv1_s5"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s5"
  type: "ReLU"
  bottom: "conv1_s5"
  top: "conv1_s5"
}
layer {
  name: "pool1_s5"
  type: "Pooling"
  bottom: "conv1_s5"
  top: "pool1_s5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s5"
  type: "LRN"
  bottom: "pool1_s5"
  top: "norm1_s5"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s5"
  type: "Convolution"
  bottom: "norm1_s5"
  top: "conv2_s5"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s5"
  type: "ReLU"
  bottom: "conv2_s5"
  top: "conv2_s5"
}
layer {
  name: "pool2_s5"
  type: "Pooling"
  bottom: "conv2_s5"
  top: "pool2_s5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s5"
  type: "LRN"
  bottom: "pool2_s5"
  top: "norm2_s5"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s5"
  type: "Convolution"
  bottom: "norm2_s5"
  top: "conv3_s5"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s5"
  type: "ReLU"
  bottom: "conv3_s5"
  top: "conv3_s5"
}
layer {
  name: "conv4_s5"
  type: "Convolution"
  bottom: "conv3_s5"
  top: "conv4_s5"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s5"
  type: "ReLU"
  bottom: "conv4_s5"
  top: "conv4_s5"
}
layer {
  name: "conv5_s5"
  type: "Convolution"
  bottom: "conv4_s5"
  top: "conv5_s5"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s5"
  type: "ReLU"
  bottom: "conv5_s5"
  top: "conv5_s5"
}
layer {
  name: "pool5_s5"
  type: "Pooling"
  bottom: "conv5_s5"
  top: "pool5_s5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s5"
  type: "InnerProduct"
  bottom: "pool5_s5"
  top: "fc6_s5"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s5"
  type: "ReLU"
  bottom: "fc6_s5"
  top: "fc6_s5"
}
layer {
  name: "drop6_s5"
  type: "Dropout"
  bottom: "fc6_s5"
  top: "fc6_s5"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_6  ############################################
layer {
  name: "conv1_s6"
  type: "Convolution"
  bottom: "data_6"
  top: "conv1_s6"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s6"
  type: "ReLU"
  bottom: "conv1_s6"
  top: "conv1_s6"
}
layer {
  name: "pool1_s6"
  type: "Pooling"
  bottom: "conv1_s6"
  top: "pool1_s6"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s6"
  type: "LRN"
  bottom: "pool1_s6"
  top: "norm1_s6"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s6"
  type: "Convolution"
  bottom: "norm1_s6"
  top: "conv2_s6"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s6"
  type: "ReLU"
  bottom: "conv2_s6"
  top: "conv2_s6"
}
layer {
  name: "pool2_s6"
  type: "Pooling"
  bottom: "conv2_s6"
  top: "pool2_s6"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s6"
  type: "LRN"
  bottom: "pool2_s6"
  top: "norm2_s6"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s6"
  type: "Convolution"
  bottom: "norm2_s6"
  top: "conv3_s6"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s6"
  type: "ReLU"
  bottom: "conv3_s6"
  top: "conv3_s6"
}
layer {
  name: "conv4_s6"
  type: "Convolution"
  bottom: "conv3_s6"
  top: "conv4_s6"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s6"
  type: "ReLU"
  bottom: "conv4_s6"
  top: "conv4_s6"
}
layer {
  name: "conv5_s6"
  type: "Convolution"
  bottom: "conv4_s6"
  top: "conv5_s6"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s6"
  type: "ReLU"
  bottom: "conv5_s6"
  top: "conv5_s6"
}
layer {
  name: "pool5_s6"
  type: "Pooling"
  bottom: "conv5_s6"
  top: "pool5_s6"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s6"
  type: "InnerProduct"
  bottom: "pool5_s6"
  top: "fc6_s6"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s6"
  type: "ReLU"
  bottom: "fc6_s6"
  top: "fc6_s6"
}
layer {
  name: "drop6_s6"
  type: "Dropout"
  bottom: "fc6_s6"
  top: "fc6_s6"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_7  ############################################
layer {
  name: "conv1_s7"
  type: "Convolution"
  bottom: "data_7"
  top: "conv1_s7"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s7"
  type: "ReLU"
  bottom: "conv1_s7"
  top: "conv1_s7"
}
layer {
  name: "pool1_s7"
  type: "Pooling"
  bottom: "conv1_s7"
  top: "pool1_s7"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s7"
  type: "LRN"
  bottom: "pool1_s7"
  top: "norm1_s7"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s7"
  type: "Convolution"
  bottom: "norm1_s7"
  top: "conv2_s7"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s7"
  type: "ReLU"
  bottom: "conv2_s7"
  top: "conv2_s7"
}
layer {
  name: "pool2_s7"
  type: "Pooling"
  bottom: "conv2_s7"
  top: "pool2_s7"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s7"
  type: "LRN"
  bottom: "pool2_s7"
  top: "norm2_s7"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s7"
  type: "Convolution"
  bottom: "norm2_s7"
  top: "conv3_s7"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s7"
  type: "ReLU"
  bottom: "conv3_s7"
  top: "conv3_s7"
}
layer {
  name: "conv4_s7"
  type: "Convolution"
  bottom: "conv3_s7"
  top: "conv4_s7"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s7"
  type: "ReLU"
  bottom: "conv4_s7"
  top: "conv4_s7"
}
layer {
  name: "conv5_s7"
  type: "Convolution"
  bottom: "conv4_s7"
  top: "conv5_s7"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s7"
  type: "ReLU"
  bottom: "conv5_s7"
  top: "conv5_s7"
}
layer {
  name: "pool5_s7"
  type: "Pooling"
  bottom: "conv5_s7"
  top: "pool5_s7"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s7"
  type: "InnerProduct"
  bottom: "pool5_s7"
  top: "fc6_s7"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s7"
  type: "ReLU"
  bottom: "fc6_s7"
  top: "fc6_s7"
}
layer {
  name: "drop6_s7"
  type: "Dropout"
  bottom: "fc6_s7"
  top: "fc6_s7"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_8 ############################################
layer {
  name: "conv1_s8"
  type: "Convolution"
  bottom: "data_8"
  top: "conv1_s8"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s8"
  type: "ReLU"
  bottom: "conv1_s8"
  top: "conv1_s8"
}
layer {
  name: "pool1_s8"
  type: "Pooling"
  bottom: "conv1_s8"
  top: "pool1_s8"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s8"
  type: "LRN"
  bottom: "pool1_s8"
  top: "norm1_s8"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s8"
  type: "Convolution"
  bottom: "norm1_s8"
  top: "conv2_s8"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s8"
  type: "ReLU"
  bottom: "conv2_s8"
  top: "conv2_s8"
}
layer {
  name: "pool2_s8"
  type: "Pooling"
  bottom: "conv2_s8"
  top: "pool2_s8"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s8"
  type: "LRN"
  bottom: "pool2_s8"
  top: "norm2_s8"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s8"
  type: "Convolution"
  bottom: "norm2_s8"
  top: "conv3_s8"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s8"
  type: "ReLU"
  bottom: "conv3_s8"
  top: "conv3_s8"
}
layer {
  name: "conv4_s8"
  type: "Convolution"
  bottom: "conv3_s8"
  top: "conv4_s8"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s8"
  type: "ReLU"
  bottom: "conv4_s8"
  top: "conv4_s8"
}
layer {
  name: "conv5_s8"
  type: "Convolution"
  bottom: "conv4_s8"
  top: "conv5_s8"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s8"
  type: "ReLU"
  bottom: "conv5_s8"
  top: "conv5_s8"
}
layer {
  name: "pool5_s8"
  type: "Pooling"
  bottom: "conv5_s8"
  top: "pool5_s8"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s8"
  type: "InnerProduct"
  bottom: "pool5_s8"
  top: "fc6_s8"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s8"
  type: "ReLU"
  bottom: "fc6_s8"
  top: "fc6_s8"
}
layer {
  name: "drop6_s8"
  type: "Dropout"
  bottom: "fc6_s8"
  top: "fc6_s8"
  dropout_param {
    dropout_ratio: 0.5
  }
}
####### Column_9 ############################################
layer {
  name: "conv1_s9"
  type: "Convolution"
  bottom: "data_9"
  top: "conv1_s9"
  param {
    name: "conv1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu1_s9"
  type: "ReLU"
  bottom: "conv1_s9"
  top: "conv1_s9"
}
layer {
  name: "pool1_s9"
  type: "Pooling"
  bottom: "conv1_s9"
  top: "pool1_s9"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm1_s9"
  type: "LRN"
  bottom: "pool1_s9"
  top: "norm1_s9"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv2_s9"
  type: "Convolution"
  bottom: "norm1_s9"
  top: "conv2_s9"
  param {
    name: "conv2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu2_s9"
  type: "ReLU"
  bottom: "conv2_s9"
  top: "conv2_s9"
}
layer {
  name: "pool2_s9"
  type: "Pooling"
  bottom: "conv2_s9"
  top: "pool2_s9"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2_s9"
  type: "LRN"
  bottom: "pool2_s9"
  top: "norm2_s9"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv3_s9"
  type: "Convolution"
  bottom: "norm2_s9"
  top: "conv3_s9"
  param {
    name: "conv3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu3_s9"
  type: "ReLU"
  bottom: "conv3_s9"
  top: "conv3_s9"
}
layer {
  name: "conv4_s9"
  type: "Convolution"
  bottom: "conv3_s9"
  top: "conv4_s9"
  param {
    name: "conv4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu4_s9"
  type: "ReLU"
  bottom: "conv4_s9"
  top: "conv4_s9"
}
layer {
  name: "conv5_s9"
  type: "Convolution"
  bottom: "conv4_s9"
  top: "conv5_s9"
  param {
    name: "conv5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "conv5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu5_s9"
  type: "ReLU"
  bottom: "conv5_s9"
  top: "conv5_s9"
}
layer {
  name: "pool5_s9"
  type: "Pooling"
  bottom: "conv5_s9"
  top: "pool5_s9"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_s9"
  type: "InnerProduct"
  bottom: "pool5_s9"
  top: "fc6_s9"
  param {
    name: "fc6s_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
    name: "fc6s_b"
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1024
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6_s9"
  type: "ReLU"
  bottom: "fc6_s9"
  top: "fc6_s9"
}
layer {
  name: "drop6_s9"
  type: "Dropout"
  bottom: "fc6_s9"
  top: "fc6_s9"
  dropout_param {
    dropout_ratio: 0.5
  }
}
######################################################################
layer {
  name: "concat"
  type: "Concat"
  bottom: "fc6_s1"
  bottom: "fc6_s2"
  bottom: "fc6_s3"
  bottom: "fc6_s4"
  bottom: "fc6_s5"
  bottom: "fc6_s6"
  bottom: "fc6_s7"
  bottom: "fc6_s8"
  bottom: "fc6_s9"
  top: "concat"
}

layer {
  name: "fc7"
  type: "InnerProduct"
  bottom: "concat"
  top: "fc7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 4096
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu7"
  type: "ReLU"
  bottom: "fc7"
  top: "fc7"
}
layer {
  name: "drop7"
  type: "Dropout"
  bottom: "fc7"
  top: "fc7"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc8"
  type: "InnerProduct"
  bottom: "fc7"
  top: "fc8"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 1000
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "accuracy"
  type: "Accuracy"
  bottom: "fc8"
  bottom: "label"
  top: "accuracy"
  include {
    phase: TEST
  }
}
layer {
  name: "loss"
  type: "SoftmaxWithLoss"
  bottom: "fc8"
  bottom: "label"
  top: "loss"
}