deploy.prototxt

name: "Inception21k"
layer { 
  name: "data"
  type: "Input"
  top: "data"
  input_param { shape: { dim: 10 dim: 3 dim: 224 dim: 224 } }
}

layer {
  name: "conv_conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv_conv1"
  convolution_param { 
     num_output: 96
     kernel_size: 7
     stride: 2
     pad: 3
  }
}

layer {
  name: "bn_conv1"
  type: "BatchNorm"
  bottom: "conv_conv1"
  top: "conv_conv1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_conv1"
  bottom: "conv_conv1"
  top: "conv_conv1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_conv1"
  type: "ReLU"
  bottom: "conv_conv1"
  top: "conv_conv1"
}

layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv_conv1"
  top: "pool1"
  pooling_param {
     pool: MAX
     kernel_size: 3
     stride: 2
     pad: 0
  }
}

layer {
  name: "conv_conv2red"
  type: "Convolution"
  bottom: "pool1"
  top: "conv_conv2red"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_conv2red"
  type: "BatchNorm"
  bottom: "conv_conv2red"
  top: "conv_conv2red"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_conv2red"
  bottom: "conv_conv2red"
  top: "conv_conv2red"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_conv2red"
  type: "ReLU"
  bottom: "conv_conv2red"
  top: "conv_conv2red"
}

layer {
  name: "conv_conv2"
  type: "Convolution"
  bottom: "conv_conv2red"
  top: "conv_conv2"
  convolution_param { 
     num_output: 288
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_conv2"
  type: "BatchNorm"
  bottom: "conv_conv2"
  top: "conv_conv2"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_conv2"
  bottom: "conv_conv2"
  top: "conv_conv2"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_conv2"
  type: "ReLU"
  bottom: "conv_conv2"
  top: "conv_conv2"
}

layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv_conv2"
  top: "pool2"
  pooling_param {
     pool: MAX
     kernel_size: 3
     stride: 2
     pad: 0
  }
}

layer {
  name: "conv_3a_1x1"
  type: "Convolution"
  bottom: "pool2"
  top: "conv_3a_1x1"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3a_1x1"
  type: "BatchNorm"
  bottom: "conv_3a_1x1"
  top: "conv_3a_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_1x1"
  bottom: "conv_3a_1x1"
  top: "conv_3a_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_1x1"
  type: "ReLU"
  bottom: "conv_3a_1x1"
  top: "conv_3a_1x1"
}

layer {
  name: "conv_3a_3x3_reduce"
  type: "Convolution"
  bottom: "pool2"
  top: "conv_3a_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3a_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_3a_3x3_reduce"
  top: "conv_3a_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_3x3_reduce"
  bottom: "conv_3a_3x3_reduce"
  top: "conv_3a_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_3x3_reduce"
  type: "ReLU"
  bottom: "conv_3a_3x3_reduce"
  top: "conv_3a_3x3_reduce"
}

layer {
  name: "conv_3a_3x3"
  type: "Convolution"
  bottom: "conv_3a_3x3_reduce"
  top: "conv_3a_3x3"
  convolution_param { 
     num_output: 96
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3a_3x3"
  type: "BatchNorm"
  bottom: "conv_3a_3x3"
  top: "conv_3a_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_3x3"
  bottom: "conv_3a_3x3"
  top: "conv_3a_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_3x3"
  type: "ReLU"
  bottom: "conv_3a_3x3"
  top: "conv_3a_3x3"
}

layer {
  name: "conv_3a_double_3x3_reduce"
  type: "Convolution"
  bottom: "pool2"
  top: "conv_3a_double_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3a_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_3a_double_3x3_reduce"
  top: "conv_3a_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_double_3x3_reduce"
  bottom: "conv_3a_double_3x3_reduce"
  top: "conv_3a_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_3a_double_3x3_reduce"
  top: "conv_3a_double_3x3_reduce"
}

layer {
  name: "conv_3a_double_3x3_0"
  type: "Convolution"
  bottom: "conv_3a_double_3x3_reduce"
  top: "conv_3a_double_3x3_0"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3a_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_3a_double_3x3_0"
  top: "conv_3a_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_double_3x3_0"
  bottom: "conv_3a_double_3x3_0"
  top: "conv_3a_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_double_3x3_0"
  type: "ReLU"
  bottom: "conv_3a_double_3x3_0"
  top: "conv_3a_double_3x3_0"
}

layer {
  name: "conv_3a_double_3x3_1"
  type: "Convolution"
  bottom: "conv_3a_double_3x3_0"
  top: "conv_3a_double_3x3_1"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3a_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_3a_double_3x3_1"
  top: "conv_3a_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_double_3x3_1"
  bottom: "conv_3a_double_3x3_1"
  top: "conv_3a_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_double_3x3_1"
  type: "ReLU"
  bottom: "conv_3a_double_3x3_1"
  top: "conv_3a_double_3x3_1"
}

layer {
  name: "ave_pool_3a_pool"
  type: "Pooling"
  bottom: "pool2"
  top: "ave_pool_3a_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_3a_proj"
  type: "Convolution"
  bottom: "ave_pool_3a_pool"
  top: "conv_3a_proj"
  convolution_param { 
     num_output: 48
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3a_proj"
  type: "BatchNorm"
  bottom: "conv_3a_proj"
  top: "conv_3a_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3a_proj"
  bottom: "conv_3a_proj"
  top: "conv_3a_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3a_proj"
  type: "ReLU"
  bottom: "conv_3a_proj"
  top: "conv_3a_proj"
}

layer {
  name: "ch_concat_3a_chconcat"
  type: "Concat"
  bottom: "conv_3a_1x1"
  bottom: "conv_3a_3x3"
  bottom: "conv_3a_double_3x3_1"
  bottom: "conv_3a_proj"
  top: "ch_concat_3a_chconcat"
}

layer {
  name: "conv_3b_1x1"
  type: "Convolution"
  bottom: "ch_concat_3a_chconcat"
  top: "conv_3b_1x1"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3b_1x1"
  type: "BatchNorm"
  bottom: "conv_3b_1x1"
  top: "conv_3b_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_1x1"
  bottom: "conv_3b_1x1"
  top: "conv_3b_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_1x1"
  type: "ReLU"
  bottom: "conv_3b_1x1"
  top: "conv_3b_1x1"
}

layer {
  name: "conv_3b_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_3a_chconcat"
  top: "conv_3b_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3b_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_3b_3x3_reduce"
  top: "conv_3b_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_3x3_reduce"
  bottom: "conv_3b_3x3_reduce"
  top: "conv_3b_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_3x3_reduce"
  type: "ReLU"
  bottom: "conv_3b_3x3_reduce"
  top: "conv_3b_3x3_reduce"
}

layer {
  name: "conv_3b_3x3"
  type: "Convolution"
  bottom: "conv_3b_3x3_reduce"
  top: "conv_3b_3x3"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3b_3x3"
  type: "BatchNorm"
  bottom: "conv_3b_3x3"
  top: "conv_3b_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_3x3"
  bottom: "conv_3b_3x3"
  top: "conv_3b_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_3x3"
  type: "ReLU"
  bottom: "conv_3b_3x3"
  top: "conv_3b_3x3"
}

layer {
  name: "conv_3b_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_3a_chconcat"
  top: "conv_3b_double_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3b_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_3b_double_3x3_reduce"
  top: "conv_3b_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_double_3x3_reduce"
  bottom: "conv_3b_double_3x3_reduce"
  top: "conv_3b_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_3b_double_3x3_reduce"
  top: "conv_3b_double_3x3_reduce"
}

layer {
  name: "conv_3b_double_3x3_0"
  type: "Convolution"
  bottom: "conv_3b_double_3x3_reduce"
  top: "conv_3b_double_3x3_0"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3b_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_3b_double_3x3_0"
  top: "conv_3b_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_double_3x3_0"
  bottom: "conv_3b_double_3x3_0"
  top: "conv_3b_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_double_3x3_0"
  type: "ReLU"
  bottom: "conv_3b_double_3x3_0"
  top: "conv_3b_double_3x3_0"
}

layer {
  name: "conv_3b_double_3x3_1"
  type: "Convolution"
  bottom: "conv_3b_double_3x3_0"
  top: "conv_3b_double_3x3_1"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3b_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_3b_double_3x3_1"
  top: "conv_3b_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_double_3x3_1"
  bottom: "conv_3b_double_3x3_1"
  top: "conv_3b_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_double_3x3_1"
  type: "ReLU"
  bottom: "conv_3b_double_3x3_1"
  top: "conv_3b_double_3x3_1"
}

layer {
  name: "ave_pool_3b_pool"
  type: "Pooling"
  bottom: "ch_concat_3a_chconcat"
  top: "ave_pool_3b_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_3b_proj"
  type: "Convolution"
  bottom: "ave_pool_3b_pool"
  top: "conv_3b_proj"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3b_proj"
  type: "BatchNorm"
  bottom: "conv_3b_proj"
  top: "conv_3b_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3b_proj"
  bottom: "conv_3b_proj"
  top: "conv_3b_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3b_proj"
  type: "ReLU"
  bottom: "conv_3b_proj"
  top: "conv_3b_proj"
}

layer {
  name: "ch_concat_3b_chconcat"
  type: "Concat"
  bottom: "conv_3b_1x1"
  bottom: "conv_3b_3x3"
  bottom: "conv_3b_double_3x3_1"
  bottom: "conv_3b_proj"
  top: "ch_concat_3b_chconcat"
}

layer {
  name: "conv_3c_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_3b_chconcat"
  top: "conv_3c_3x3_reduce"
  convolution_param { 
     num_output: 192
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3c_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_3c_3x3_reduce"
  top: "conv_3c_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3c_3x3_reduce"
  bottom: "conv_3c_3x3_reduce"
  top: "conv_3c_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3c_3x3_reduce"
  type: "ReLU"
  bottom: "conv_3c_3x3_reduce"
  top: "conv_3c_3x3_reduce"
}

layer {
  name: "conv_3c_3x3"
  type: "Convolution"
  bottom: "conv_3c_3x3_reduce"
  top: "conv_3c_3x3"
  convolution_param { 
     num_output: 240
     kernel_size: 3
     stride: 2
     pad: 1
  }
}

layer {
  name: "bn_3c_3x3"
  type: "BatchNorm"
  bottom: "conv_3c_3x3"
  top: "conv_3c_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3c_3x3"
  bottom: "conv_3c_3x3"
  top: "conv_3c_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3c_3x3"
  type: "ReLU"
  bottom: "conv_3c_3x3"
  top: "conv_3c_3x3"
}

layer {
  name: "conv_3c_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_3b_chconcat"
  top: "conv_3c_double_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_3c_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_3c_double_3x3_reduce"
  top: "conv_3c_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3c_double_3x3_reduce"
  bottom: "conv_3c_double_3x3_reduce"
  top: "conv_3c_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3c_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_3c_double_3x3_reduce"
  top: "conv_3c_double_3x3_reduce"
}

layer {
  name: "conv_3c_double_3x3_0"
  type: "Convolution"
  bottom: "conv_3c_double_3x3_reduce"
  top: "conv_3c_double_3x3_0"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_3c_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_3c_double_3x3_0"
  top: "conv_3c_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3c_double_3x3_0"
  bottom: "conv_3c_double_3x3_0"
  top: "conv_3c_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3c_double_3x3_0"
  type: "ReLU"
  bottom: "conv_3c_double_3x3_0"
  top: "conv_3c_double_3x3_0"
}

layer {
  name: "conv_3c_double_3x3_1"
  type: "Convolution"
  bottom: "conv_3c_double_3x3_0"
  top: "conv_3c_double_3x3_1"
  convolution_param { 
     num_output: 144
     kernel_size: 3
     stride: 2
     pad: 1
  }
}

layer {
  name: "bn_3c_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_3c_double_3x3_1"
  top: "conv_3c_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_3c_double_3x3_1"
  bottom: "conv_3c_double_3x3_1"
  top: "conv_3c_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_3c_double_3x3_1"
  type: "ReLU"
  bottom: "conv_3c_double_3x3_1"
  top: "conv_3c_double_3x3_1"
}

layer {
  name: "max_pool_3c_pool"
  type: "Pooling"
  bottom: "ch_concat_3b_chconcat"
  top: "max_pool_3c_pool"
  pooling_param {
     pool: MAX
     kernel_size: 3
     stride: 2
     pad: 0
  }
}

layer {
  name: "ch_concat_3c_chconcat"
  type: "Concat"
  bottom: "conv_3c_3x3"
  bottom: "conv_3c_double_3x3_1"
  bottom: "max_pool_3c_pool"
  top: "ch_concat_3c_chconcat"
}

layer {
  name: "conv_4a_1x1"
  type: "Convolution"
  bottom: "ch_concat_3c_chconcat"
  top: "conv_4a_1x1"
  convolution_param { 
     num_output: 224
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4a_1x1"
  type: "BatchNorm"
  bottom: "conv_4a_1x1"
  top: "conv_4a_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_1x1"
  bottom: "conv_4a_1x1"
  top: "conv_4a_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_1x1"
  type: "ReLU"
  bottom: "conv_4a_1x1"
  top: "conv_4a_1x1"
}

layer {
  name: "conv_4a_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_3c_chconcat"
  top: "conv_4a_3x3_reduce"
  convolution_param { 
     num_output: 64
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4a_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4a_3x3_reduce"
  top: "conv_4a_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_3x3_reduce"
  bottom: "conv_4a_3x3_reduce"
  top: "conv_4a_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4a_3x3_reduce"
  top: "conv_4a_3x3_reduce"
}

layer {
  name: "conv_4a_3x3"
  type: "Convolution"
  bottom: "conv_4a_3x3_reduce"
  top: "conv_4a_3x3"
  convolution_param { 
     num_output: 96
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4a_3x3"
  type: "BatchNorm"
  bottom: "conv_4a_3x3"
  top: "conv_4a_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_3x3"
  bottom: "conv_4a_3x3"
  top: "conv_4a_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_3x3"
  type: "ReLU"
  bottom: "conv_4a_3x3"
  top: "conv_4a_3x3"
}

layer {
  name: "conv_4a_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_3c_chconcat"
  top: "conv_4a_double_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4a_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4a_double_3x3_reduce"
  top: "conv_4a_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_double_3x3_reduce"
  bottom: "conv_4a_double_3x3_reduce"
  top: "conv_4a_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4a_double_3x3_reduce"
  top: "conv_4a_double_3x3_reduce"
}

layer {
  name: "conv_4a_double_3x3_0"
  type: "Convolution"
  bottom: "conv_4a_double_3x3_reduce"
  top: "conv_4a_double_3x3_0"
  convolution_param { 
     num_output: 128
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4a_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_4a_double_3x3_0"
  top: "conv_4a_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_double_3x3_0"
  bottom: "conv_4a_double_3x3_0"
  top: "conv_4a_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_double_3x3_0"
  type: "ReLU"
  bottom: "conv_4a_double_3x3_0"
  top: "conv_4a_double_3x3_0"
}

layer {
  name: "conv_4a_double_3x3_1"
  type: "Convolution"
  bottom: "conv_4a_double_3x3_0"
  top: "conv_4a_double_3x3_1"
  convolution_param { 
     num_output: 128
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4a_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_4a_double_3x3_1"
  top: "conv_4a_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_double_3x3_1"
  bottom: "conv_4a_double_3x3_1"
  top: "conv_4a_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_double_3x3_1"
  type: "ReLU"
  bottom: "conv_4a_double_3x3_1"
  top: "conv_4a_double_3x3_1"
}

layer {
  name: "ave_pool_4a_pool"
  type: "Pooling"
  bottom: "ch_concat_3c_chconcat"
  top: "ave_pool_4a_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_4a_proj"
  type: "Convolution"
  bottom: "ave_pool_4a_pool"
  top: "conv_4a_proj"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4a_proj"
  type: "BatchNorm"
  bottom: "conv_4a_proj"
  top: "conv_4a_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4a_proj"
  bottom: "conv_4a_proj"
  top: "conv_4a_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4a_proj"
  type: "ReLU"
  bottom: "conv_4a_proj"
  top: "conv_4a_proj"
}

layer {
  name: "ch_concat_4a_chconcat"
  type: "Concat"
  bottom: "conv_4a_1x1"
  bottom: "conv_4a_3x3"
  bottom: "conv_4a_double_3x3_1"
  bottom: "conv_4a_proj"
  top: "ch_concat_4a_chconcat"
}

layer {
  name: "conv_4b_1x1"
  type: "Convolution"
  bottom: "ch_concat_4a_chconcat"
  top: "conv_4b_1x1"
  convolution_param { 
     num_output: 192
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4b_1x1"
  type: "BatchNorm"
  bottom: "conv_4b_1x1"
  top: "conv_4b_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_1x1"
  bottom: "conv_4b_1x1"
  top: "conv_4b_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_1x1"
  type: "ReLU"
  bottom: "conv_4b_1x1"
  top: "conv_4b_1x1"
}

layer {
  name: "conv_4b_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4a_chconcat"
  top: "conv_4b_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4b_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4b_3x3_reduce"
  top: "conv_4b_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_3x3_reduce"
  bottom: "conv_4b_3x3_reduce"
  top: "conv_4b_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4b_3x3_reduce"
  top: "conv_4b_3x3_reduce"
}

layer {
  name: "conv_4b_3x3"
  type: "Convolution"
  bottom: "conv_4b_3x3_reduce"
  top: "conv_4b_3x3"
  convolution_param { 
     num_output: 128
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4b_3x3"
  type: "BatchNorm"
  bottom: "conv_4b_3x3"
  top: "conv_4b_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_3x3"
  bottom: "conv_4b_3x3"
  top: "conv_4b_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_3x3"
  type: "ReLU"
  bottom: "conv_4b_3x3"
  top: "conv_4b_3x3"
}

layer {
  name: "conv_4b_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4a_chconcat"
  top: "conv_4b_double_3x3_reduce"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4b_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4b_double_3x3_reduce"
  top: "conv_4b_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_double_3x3_reduce"
  bottom: "conv_4b_double_3x3_reduce"
  top: "conv_4b_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4b_double_3x3_reduce"
  top: "conv_4b_double_3x3_reduce"
}

layer {
  name: "conv_4b_double_3x3_0"
  type: "Convolution"
  bottom: "conv_4b_double_3x3_reduce"
  top: "conv_4b_double_3x3_0"
  convolution_param { 
     num_output: 128
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4b_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_4b_double_3x3_0"
  top: "conv_4b_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_double_3x3_0"
  bottom: "conv_4b_double_3x3_0"
  top: "conv_4b_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_double_3x3_0"
  type: "ReLU"
  bottom: "conv_4b_double_3x3_0"
  top: "conv_4b_double_3x3_0"
}

layer {
  name: "conv_4b_double_3x3_1"
  type: "Convolution"
  bottom: "conv_4b_double_3x3_0"
  top: "conv_4b_double_3x3_1"
  convolution_param { 
     num_output: 128
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4b_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_4b_double_3x3_1"
  top: "conv_4b_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_double_3x3_1"
  bottom: "conv_4b_double_3x3_1"
  top: "conv_4b_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_double_3x3_1"
  type: "ReLU"
  bottom: "conv_4b_double_3x3_1"
  top: "conv_4b_double_3x3_1"
}

layer {
  name: "ave_pool_4b_pool"
  type: "Pooling"
  bottom: "ch_concat_4a_chconcat"
  top: "ave_pool_4b_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_4b_proj"
  type: "Convolution"
  bottom: "ave_pool_4b_pool"
  top: "conv_4b_proj"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4b_proj"
  type: "BatchNorm"
  bottom: "conv_4b_proj"
  top: "conv_4b_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4b_proj"
  bottom: "conv_4b_proj"
  top: "conv_4b_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4b_proj"
  type: "ReLU"
  bottom: "conv_4b_proj"
  top: "conv_4b_proj"
}

layer {
  name: "ch_concat_4b_chconcat"
  type: "Concat"
  bottom: "conv_4b_1x1"
  bottom: "conv_4b_3x3"
  bottom: "conv_4b_double_3x3_1"
  bottom: "conv_4b_proj"
  top: "ch_concat_4b_chconcat"
}

layer {
  name: "conv_4c_1x1"
  type: "Convolution"
  bottom: "ch_concat_4b_chconcat"
  top: "conv_4c_1x1"
  convolution_param { 
     num_output: 160
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4c_1x1"
  type: "BatchNorm"
  bottom: "conv_4c_1x1"
  top: "conv_4c_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_1x1"
  bottom: "conv_4c_1x1"
  top: "conv_4c_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_1x1"
  type: "ReLU"
  bottom: "conv_4c_1x1"
  top: "conv_4c_1x1"
}

layer {
  name: "conv_4c_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4b_chconcat"
  top: "conv_4c_3x3_reduce"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4c_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4c_3x3_reduce"
  top: "conv_4c_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_3x3_reduce"
  bottom: "conv_4c_3x3_reduce"
  top: "conv_4c_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4c_3x3_reduce"
  top: "conv_4c_3x3_reduce"
}

layer {
  name: "conv_4c_3x3"
  type: "Convolution"
  bottom: "conv_4c_3x3_reduce"
  top: "conv_4c_3x3"
  convolution_param { 
     num_output: 160
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4c_3x3"
  type: "BatchNorm"
  bottom: "conv_4c_3x3"
  top: "conv_4c_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_3x3"
  bottom: "conv_4c_3x3"
  top: "conv_4c_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_3x3"
  type: "ReLU"
  bottom: "conv_4c_3x3"
  top: "conv_4c_3x3"
}

layer {
  name: "conv_4c_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4b_chconcat"
  top: "conv_4c_double_3x3_reduce"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4c_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4c_double_3x3_reduce"
  top: "conv_4c_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_double_3x3_reduce"
  bottom: "conv_4c_double_3x3_reduce"
  top: "conv_4c_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4c_double_3x3_reduce"
  top: "conv_4c_double_3x3_reduce"
}

layer {
  name: "conv_4c_double_3x3_0"
  type: "Convolution"
  bottom: "conv_4c_double_3x3_reduce"
  top: "conv_4c_double_3x3_0"
  convolution_param { 
     num_output: 160
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4c_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_4c_double_3x3_0"
  top: "conv_4c_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_double_3x3_0"
  bottom: "conv_4c_double_3x3_0"
  top: "conv_4c_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_double_3x3_0"
  type: "ReLU"
  bottom: "conv_4c_double_3x3_0"
  top: "conv_4c_double_3x3_0"
}

layer {
  name: "conv_4c_double_3x3_1"
  type: "Convolution"
  bottom: "conv_4c_double_3x3_0"
  top: "conv_4c_double_3x3_1"
  convolution_param { 
     num_output: 160
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4c_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_4c_double_3x3_1"
  top: "conv_4c_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_double_3x3_1"
  bottom: "conv_4c_double_3x3_1"
  top: "conv_4c_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_double_3x3_1"
  type: "ReLU"
  bottom: "conv_4c_double_3x3_1"
  top: "conv_4c_double_3x3_1"
}

layer {
  name: "ave_pool_4c_pool"
  type: "Pooling"
  bottom: "ch_concat_4b_chconcat"
  top: "ave_pool_4c_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_4c_proj"
  type: "Convolution"
  bottom: "ave_pool_4c_pool"
  top: "conv_4c_proj"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4c_proj"
  type: "BatchNorm"
  bottom: "conv_4c_proj"
  top: "conv_4c_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4c_proj"
  bottom: "conv_4c_proj"
  top: "conv_4c_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4c_proj"
  type: "ReLU"
  bottom: "conv_4c_proj"
  top: "conv_4c_proj"
}

layer {
  name: "ch_concat_4c_chconcat"
  type: "Concat"
  bottom: "conv_4c_1x1"
  bottom: "conv_4c_3x3"
  bottom: "conv_4c_double_3x3_1"
  bottom: "conv_4c_proj"
  top: "ch_concat_4c_chconcat"
}

layer {
  name: "conv_4d_1x1"
  type: "Convolution"
  bottom: "ch_concat_4c_chconcat"
  top: "conv_4d_1x1"
  convolution_param { 
     num_output: 96
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4d_1x1"
  type: "BatchNorm"
  bottom: "conv_4d_1x1"
  top: "conv_4d_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_1x1"
  bottom: "conv_4d_1x1"
  top: "conv_4d_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_1x1"
  type: "ReLU"
  bottom: "conv_4d_1x1"
  top: "conv_4d_1x1"
}

layer {
  name: "conv_4d_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4c_chconcat"
  top: "conv_4d_3x3_reduce"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4d_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4d_3x3_reduce"
  top: "conv_4d_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_3x3_reduce"
  bottom: "conv_4d_3x3_reduce"
  top: "conv_4d_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4d_3x3_reduce"
  top: "conv_4d_3x3_reduce"
}

layer {
  name: "conv_4d_3x3"
  type: "Convolution"
  bottom: "conv_4d_3x3_reduce"
  top: "conv_4d_3x3"
  convolution_param { 
     num_output: 192
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4d_3x3"
  type: "BatchNorm"
  bottom: "conv_4d_3x3"
  top: "conv_4d_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_3x3"
  bottom: "conv_4d_3x3"
  top: "conv_4d_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_3x3"
  type: "ReLU"
  bottom: "conv_4d_3x3"
  top: "conv_4d_3x3"
}

layer {
  name: "conv_4d_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4c_chconcat"
  top: "conv_4d_double_3x3_reduce"
  convolution_param { 
     num_output: 160
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4d_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4d_double_3x3_reduce"
  top: "conv_4d_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_double_3x3_reduce"
  bottom: "conv_4d_double_3x3_reduce"
  top: "conv_4d_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4d_double_3x3_reduce"
  top: "conv_4d_double_3x3_reduce"
}

layer {
  name: "conv_4d_double_3x3_0"
  type: "Convolution"
  bottom: "conv_4d_double_3x3_reduce"
  top: "conv_4d_double_3x3_0"
  convolution_param { 
     num_output: 96
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4d_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_4d_double_3x3_0"
  top: "conv_4d_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_double_3x3_0"
  bottom: "conv_4d_double_3x3_0"
  top: "conv_4d_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_double_3x3_0"
  type: "ReLU"
  bottom: "conv_4d_double_3x3_0"
  top: "conv_4d_double_3x3_0"
}

layer {
  name: "conv_4d_double_3x3_1"
  type: "Convolution"
  bottom: "conv_4d_double_3x3_0"
  top: "conv_4d_double_3x3_1"
  convolution_param { 
     num_output: 96
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4d_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_4d_double_3x3_1"
  top: "conv_4d_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_double_3x3_1"
  bottom: "conv_4d_double_3x3_1"
  top: "conv_4d_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_double_3x3_1"
  type: "ReLU"
  bottom: "conv_4d_double_3x3_1"
  top: "conv_4d_double_3x3_1"
}

layer {
  name: "ave_pool_4d_pool"
  type: "Pooling"
  bottom: "ch_concat_4c_chconcat"
  top: "ave_pool_4d_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_4d_proj"
  type: "Convolution"
  bottom: "ave_pool_4d_pool"
  top: "conv_4d_proj"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4d_proj"
  type: "BatchNorm"
  bottom: "conv_4d_proj"
  top: "conv_4d_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4d_proj"
  bottom: "conv_4d_proj"
  top: "conv_4d_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4d_proj"
  type: "ReLU"
  bottom: "conv_4d_proj"
  top: "conv_4d_proj"
}

layer {
  name: "ch_concat_4d_chconcat"
  type: "Concat"
  bottom: "conv_4d_1x1"
  bottom: "conv_4d_3x3"
  bottom: "conv_4d_double_3x3_1"
  bottom: "conv_4d_proj"
  top: "ch_concat_4d_chconcat"
}

layer {
  name: "conv_4e_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4d_chconcat"
  top: "conv_4e_3x3_reduce"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4e_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4e_3x3_reduce"
  top: "conv_4e_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4e_3x3_reduce"
  bottom: "conv_4e_3x3_reduce"
  top: "conv_4e_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4e_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4e_3x3_reduce"
  top: "conv_4e_3x3_reduce"
}

layer {
  name: "conv_4e_3x3"
  type: "Convolution"
  bottom: "conv_4e_3x3_reduce"
  top: "conv_4e_3x3"
  convolution_param { 
     num_output: 192
     kernel_size: 3
     stride: 2
     pad: 1
  }
}

layer {
  name: "bn_4e_3x3"
  type: "BatchNorm"
  bottom: "conv_4e_3x3"
  top: "conv_4e_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4e_3x3"
  bottom: "conv_4e_3x3"
  top: "conv_4e_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4e_3x3"
  type: "ReLU"
  bottom: "conv_4e_3x3"
  top: "conv_4e_3x3"
}

layer {
  name: "conv_4e_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4d_chconcat"
  top: "conv_4e_double_3x3_reduce"
  convolution_param { 
     num_output: 192
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_4e_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_4e_double_3x3_reduce"
  top: "conv_4e_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4e_double_3x3_reduce"
  bottom: "conv_4e_double_3x3_reduce"
  top: "conv_4e_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4e_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_4e_double_3x3_reduce"
  top: "conv_4e_double_3x3_reduce"
}

layer {
  name: "conv_4e_double_3x3_0"
  type: "Convolution"
  bottom: "conv_4e_double_3x3_reduce"
  top: "conv_4e_double_3x3_0"
  convolution_param { 
     num_output: 256
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_4e_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_4e_double_3x3_0"
  top: "conv_4e_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4e_double_3x3_0"
  bottom: "conv_4e_double_3x3_0"
  top: "conv_4e_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4e_double_3x3_0"
  type: "ReLU"
  bottom: "conv_4e_double_3x3_0"
  top: "conv_4e_double_3x3_0"
}

layer {
  name: "conv_4e_double_3x3_1"
  type: "Convolution"
  bottom: "conv_4e_double_3x3_0"
  top: "conv_4e_double_3x3_1"
  convolution_param { 
     num_output: 256
     kernel_size: 3
     stride: 2
     pad: 1
  }
}

layer {
  name: "bn_4e_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_4e_double_3x3_1"
  top: "conv_4e_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_4e_double_3x3_1"
  bottom: "conv_4e_double_3x3_1"
  top: "conv_4e_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_4e_double_3x3_1"
  type: "ReLU"
  bottom: "conv_4e_double_3x3_1"
  top: "conv_4e_double_3x3_1"
}

layer {
  name: "max_pool_4e_pool"
  type: "Pooling"
  bottom: "ch_concat_4d_chconcat"
  top: "max_pool_4e_pool"
  pooling_param {
     pool: MAX
     kernel_size: 3
     stride: 2
     pad: 0
  }
}

layer {
  name: "ch_concat_4e_chconcat"
  type: "Concat"
  bottom: "conv_4e_3x3"
  bottom: "conv_4e_double_3x3_1"
  bottom: "max_pool_4e_pool"
  top: "ch_concat_4e_chconcat"
}

layer {
  name: "conv_5a_1x1"
  type: "Convolution"
  bottom: "ch_concat_4e_chconcat"
  top: "conv_5a_1x1"
  convolution_param { 
     num_output: 352
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5a_1x1"
  type: "BatchNorm"
  bottom: "conv_5a_1x1"
  top: "conv_5a_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_1x1"
  bottom: "conv_5a_1x1"
  top: "conv_5a_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_1x1"
  type: "ReLU"
  bottom: "conv_5a_1x1"
  top: "conv_5a_1x1"
}

layer {
  name: "conv_5a_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4e_chconcat"
  top: "conv_5a_3x3_reduce"
  convolution_param { 
     num_output: 192
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5a_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_5a_3x3_reduce"
  top: "conv_5a_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_3x3_reduce"
  bottom: "conv_5a_3x3_reduce"
  top: "conv_5a_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_3x3_reduce"
  type: "ReLU"
  bottom: "conv_5a_3x3_reduce"
  top: "conv_5a_3x3_reduce"
}

layer {
  name: "conv_5a_3x3"
  type: "Convolution"
  bottom: "conv_5a_3x3_reduce"
  top: "conv_5a_3x3"
  convolution_param { 
     num_output: 320
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_5a_3x3"
  type: "BatchNorm"
  bottom: "conv_5a_3x3"
  top: "conv_5a_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_3x3"
  bottom: "conv_5a_3x3"
  top: "conv_5a_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_3x3"
  type: "ReLU"
  bottom: "conv_5a_3x3"
  top: "conv_5a_3x3"
}

layer {
  name: "conv_5a_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_4e_chconcat"
  top: "conv_5a_double_3x3_reduce"
  convolution_param { 
     num_output: 160
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5a_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_5a_double_3x3_reduce"
  top: "conv_5a_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_double_3x3_reduce"
  bottom: "conv_5a_double_3x3_reduce"
  top: "conv_5a_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_5a_double_3x3_reduce"
  top: "conv_5a_double_3x3_reduce"
}

layer {
  name: "conv_5a_double_3x3_0"
  type: "Convolution"
  bottom: "conv_5a_double_3x3_reduce"
  top: "conv_5a_double_3x3_0"
  convolution_param { 
     num_output: 224
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_5a_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_5a_double_3x3_0"
  top: "conv_5a_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_double_3x3_0"
  bottom: "conv_5a_double_3x3_0"
  top: "conv_5a_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_double_3x3_0"
  type: "ReLU"
  bottom: "conv_5a_double_3x3_0"
  top: "conv_5a_double_3x3_0"
}

layer {
  name: "conv_5a_double_3x3_1"
  type: "Convolution"
  bottom: "conv_5a_double_3x3_0"
  top: "conv_5a_double_3x3_1"
  convolution_param { 
     num_output: 224
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_5a_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_5a_double_3x3_1"
  top: "conv_5a_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_double_3x3_1"
  bottom: "conv_5a_double_3x3_1"
  top: "conv_5a_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_double_3x3_1"
  type: "ReLU"
  bottom: "conv_5a_double_3x3_1"
  top: "conv_5a_double_3x3_1"
}

layer {
  name: "ave_pool_5a_pool"
  type: "Pooling"
  bottom: "ch_concat_4e_chconcat"
  top: "ave_pool_5a_pool"
  pooling_param {
     pool: AVE
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_5a_proj"
  type: "Convolution"
  bottom: "ave_pool_5a_pool"
  top: "conv_5a_proj"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5a_proj"
  type: "BatchNorm"
  bottom: "conv_5a_proj"
  top: "conv_5a_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5a_proj"
  bottom: "conv_5a_proj"
  top: "conv_5a_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5a_proj"
  type: "ReLU"
  bottom: "conv_5a_proj"
  top: "conv_5a_proj"
}

layer {
  name: "ch_concat_5a_chconcat"
  type: "Concat"
  bottom: "conv_5a_1x1"
  bottom: "conv_5a_3x3"
  bottom: "conv_5a_double_3x3_1"
  bottom: "conv_5a_proj"
  top: "ch_concat_5a_chconcat"
}

layer {
  name: "conv_5b_1x1"
  type: "Convolution"
  bottom: "ch_concat_5a_chconcat"
  top: "conv_5b_1x1"
  convolution_param { 
     num_output: 352
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5b_1x1"
  type: "BatchNorm"
  bottom: "conv_5b_1x1"
  top: "conv_5b_1x1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_1x1"
  bottom: "conv_5b_1x1"
  top: "conv_5b_1x1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_1x1"
  type: "ReLU"
  bottom: "conv_5b_1x1"
  top: "conv_5b_1x1"
}

layer {
  name: "conv_5b_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_5a_chconcat"
  top: "conv_5b_3x3_reduce"
  convolution_param { 
     num_output: 192
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5b_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_5b_3x3_reduce"
  top: "conv_5b_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_3x3_reduce"
  bottom: "conv_5b_3x3_reduce"
  top: "conv_5b_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_3x3_reduce"
  type: "ReLU"
  bottom: "conv_5b_3x3_reduce"
  top: "conv_5b_3x3_reduce"
}

layer {
  name: "conv_5b_3x3"
  type: "Convolution"
  bottom: "conv_5b_3x3_reduce"
  top: "conv_5b_3x3"
  convolution_param { 
     num_output: 320
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_5b_3x3"
  type: "BatchNorm"
  bottom: "conv_5b_3x3"
  top: "conv_5b_3x3"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_3x3"
  bottom: "conv_5b_3x3"
  top: "conv_5b_3x3"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_3x3"
  type: "ReLU"
  bottom: "conv_5b_3x3"
  top: "conv_5b_3x3"
}

layer {
  name: "conv_5b_double_3x3_reduce"
  type: "Convolution"
  bottom: "ch_concat_5a_chconcat"
  top: "conv_5b_double_3x3_reduce"
  convolution_param { 
     num_output: 192
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5b_double_3x3_reduce"
  type: "BatchNorm"
  bottom: "conv_5b_double_3x3_reduce"
  top: "conv_5b_double_3x3_reduce"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_double_3x3_reduce"
  bottom: "conv_5b_double_3x3_reduce"
  top: "conv_5b_double_3x3_reduce"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_double_3x3_reduce"
  type: "ReLU"
  bottom: "conv_5b_double_3x3_reduce"
  top: "conv_5b_double_3x3_reduce"
}

layer {
  name: "conv_5b_double_3x3_0"
  type: "Convolution"
  bottom: "conv_5b_double_3x3_reduce"
  top: "conv_5b_double_3x3_0"
  convolution_param { 
     num_output: 224
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_5b_double_3x3_0"
  type: "BatchNorm"
  bottom: "conv_5b_double_3x3_0"
  top: "conv_5b_double_3x3_0"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_double_3x3_0"
  bottom: "conv_5b_double_3x3_0"
  top: "conv_5b_double_3x3_0"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_double_3x3_0"
  type: "ReLU"
  bottom: "conv_5b_double_3x3_0"
  top: "conv_5b_double_3x3_0"
}

layer {
  name: "conv_5b_double_3x3_1"
  type: "Convolution"
  bottom: "conv_5b_double_3x3_0"
  top: "conv_5b_double_3x3_1"
  convolution_param { 
     num_output: 224
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "bn_5b_double_3x3_1"
  type: "BatchNorm"
  bottom: "conv_5b_double_3x3_1"
  top: "conv_5b_double_3x3_1"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_double_3x3_1"
  bottom: "conv_5b_double_3x3_1"
  top: "conv_5b_double_3x3_1"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_double_3x3_1"
  type: "ReLU"
  bottom: "conv_5b_double_3x3_1"
  top: "conv_5b_double_3x3_1"
}

layer {
  name: "max_pool_5b_pool"
  type: "Pooling"
  bottom: "ch_concat_5a_chconcat"
  top: "max_pool_5b_pool"
  pooling_param {
     pool: MAX
     kernel_size: 3
     stride: 1
     pad: 1
  }
}

layer {
  name: "conv_5b_proj"
  type: "Convolution"
  bottom: "max_pool_5b_pool"
  top: "conv_5b_proj"
  convolution_param { 
     num_output: 128
     kernel_size: 1
     stride: 1
     pad: 0
  }
}

layer {
  name: "bn_5b_proj"
  type: "BatchNorm"
  bottom: "conv_5b_proj"
  top: "conv_5b_proj"
  batch_norm_param {
    use_global_stats: true
  }
}

layer {
  name: "scale_conv_5b_proj"
  bottom: "conv_5b_proj"
  top: "conv_5b_proj"
  type: "Scale"
  scale_param {
        bias_term: true
  }
}

layer {
  name: "relu_5b_proj"
  type: "ReLU"
  bottom: "conv_5b_proj"
  top: "conv_5b_proj"
}

layer {
  name: "ch_concat_5b_chconcat"
  type: "Concat"
  bottom: "conv_5b_1x1"
  bottom: "conv_5b_3x3"
  bottom: "conv_5b_double_3x3_1"
  bottom: "conv_5b_proj"
  top: "ch_concat_5b_chconcat"
}

layer {
  name: "global_pool"
  type: "Pooling"
  bottom: "ch_concat_5b_chconcat"
  top: "global_pool"
  pooling_param {
     global_pooling : true
     pool: AVE
  }
}

layer {
  name: "fc1"
  type: "InnerProduct"
  bottom: "global_pool"
  top: "fc1"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 21841
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "softmax"
  type: "Softmax"
  bottom: "fc1"
  top: "softmax"
}