# This is basically the same as the default linear test trainer
# but here we have removed python layers so we can run in a
# multi-gpu system. This one also works with CaffeNet variant
# of AlexNet. The python layers are only for display, so we 
# really don't miss them. You will need to adjust the batch
# size for how many GPU's you are using. 

name: "CaffeNet"
# ************************
# ***** DATA LAYER *******
# ************************
# force_backward: True

# train data
layer {
  top: "data"
  top: "label"
  name: "data"
  type: "ImageData"
  image_data_param {
    source: "/tmp/train_list.txt"
    # 1 GPU : Iter Size 4 (effective 256 batch size)     
    #batch_size: 64 
    # 2 GPU : Iter Size 1
    #batch_size: 128
    # 4 GPU : Iter Size 1
    batch_size: 64
    shuffle: true
  }
  transform_param {
    mirror: true
    crop_size: 227
    mean_value: 104
    mean_value: 117
    mean_value: 123
  }
  include {
    phase: TRAIN
  }
}

# test data
layer {
  top: "data"
  top: "label"
  name: "data"
  type: "ImageData"
  image_data_param {
    source: "/tmp/test_list.txt"
    # Suggested 25
    batch_size: 25
  }
  transform_param {
    mirror: false
    crop_size: 227
    mean_value: 104
    mean_value: 117
    mean_value: 123
  }
  include {
    phase: TEST
  }  
}


# ================================================================================================
# ================================================================================================
# ======== NETWORK 1
# ================================================================================================
# ================================================================================================

layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv1_w"
  }
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv1_b"
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 4
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1"
  type: "ReLU"
  bottom: "conv1"
  top: "conv1"
}

layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}

layer {
  name: "norm1"
  type: "LRN"
  bottom: "pool1"
  top: "norm1"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}

layer {
  name: "conv2"
  type: "Convolution"
  bottom: "norm1"
  top: "conv2"  
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv2_w"
  }
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv2_b"
  }
  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"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}

layer {
  name: "pool2"
  type: "Pooling"
  bottom: "conv2"
  top: "pool2"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm2"
  type: "LRN"
  bottom: "pool2"
  top: "norm2"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}

layer {
  name: "conv3"
  type: "Convolution"
  bottom: "norm2"
  top: "conv3" 
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv3_w"
  }
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv3_b"
  }
  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"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}

layer {
  name: "conv4"
  type: "Convolution"
  bottom: "conv3"
  top: "conv4" 
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv4_w"
  }
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv4_b"
  }
  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"
  type: "ReLU"
  bottom: "conv4"
  top: "conv4"
}

layer {
  name: "conv5"
  type: "Convolution"
  bottom: "conv4"
  top: "conv5"  
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv5_w"
  }
  param {
    lr_mult: 0
    decay_mult: 0
    name: "conv5_b"
  }
  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"
  type: "ReLU"
  bottom: "conv5"
  top: "conv5"
}

layer {
  name: "pool5"
  type: "Pooling"
  bottom: "conv5"
  top: "pool5"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}


# **************************
# ***** POOLING LAYERS *****
# **************************
layer {
  name: "conv1_pool"
  type: "Resize"
  bottom: "conv1"
  top: "conv1_pool"
  resize_param {
    height: 10
    width: 10
  }
}
layer {
  name: "pool1_pool"
  type: "Resize"
  bottom: "pool1"
  top: "pool1_pool"
  resize_param {
    height: 10
    width: 10
  }
}
layer {
  name: "conv2_pool"
  type: "Resize"
  bottom: "conv2"
  top: "conv2_pool"
  resize_param {
    height: 6
    width: 6
  }
}
layer {
  name: "pool2_pool"
  type: "Resize"
  bottom: "pool2"
  top: "pool2_pool"
  resize_param {
    height: 6
    width: 6
  }
}
layer {
  name: "conv3_pool"
  type: "Resize"
  bottom: "conv3"
  top: "conv3_pool"
  resize_param {
    height: 5
    width: 5
  }
}
layer {
  name: "conv4_pool"
  type: "Resize"
  bottom: "conv4"
  top: "conv4_pool"
  resize_param {
    height: 5
    width: 5
  }
}
layer {
  name: "conv5_pool"
  type: "Resize"
  bottom: "conv5"
  top: "conv5_pool"
  resize_param {
    height: 6
    width: 6
  }
}

# ********************************
# ***** NORMALIZATION LAYERS *****
# ********************************
layer {
  name: "conv1_norm"
  type: "BatchNorm"
  bottom: "conv1_pool"
  top: "conv1_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "pool1_norm"
  type: "BatchNorm"
  bottom: "pool1_pool"
  top: "pool1_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "conv2_norm"
  type: "BatchNorm"
  bottom: "conv2_pool"
  top: "conv2_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "pool2_norm"
  type: "BatchNorm"
  bottom: "pool2_pool"
  top: "pool2_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "conv3_norm"
  type: "BatchNorm"
  bottom: "conv3_pool"
  top: "conv3_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "conv4_norm"
  type: "BatchNorm"
  bottom: "conv4_pool"
  top: "conv4_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "conv5_norm"
  type: "BatchNorm"
  bottom: "conv5_pool"
  top: "conv5_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}
layer {
  name: "pool5_norm"
  type: "BatchNorm"
  bottom: "pool5"
  top: "pool5_norm"
  batch_norm_param{ }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
}

# *********************************
# ***** CLASSIFICATION LAYERS *****
# *********************************
layer {
  name: "conv1_cl"
  type: "InnerProduct"
  bottom: "conv1_norm"
  top: "conv1_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "pool1_cl"
  type: "InnerProduct"
  bottom: "pool1_norm"
  top: "pool1_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "conv2_cl"
  type: "InnerProduct"
  bottom: "conv2_norm"
  top: "conv2_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "pool2_cl"
  type: "InnerProduct"
  bottom: "pool2_norm"
  top: "pool2_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "conv3_cl"
  type: "InnerProduct"
  bottom: "conv3_norm"
  top: "conv3_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "conv4_cl"
  type: "InnerProduct"
  bottom: "conv4_norm"
  top: "conv4_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "conv5_cl"
  type: "InnerProduct"
  bottom: "conv5_norm"
  top: "conv5_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}
layer {
  name: "pool5_cl"
  type: "InnerProduct"
  bottom: "pool5_norm"
  top: "pool5_cl"
  param {    lr_mult: 1    decay_mult: 1  }
  param {    lr_mult: 2    decay_mult: 0  }
  inner_product_param {
    num_output: 1000
  }
}

# ***********************
# ***** LOSS LAYERS *****
# ***********************
layer {
  name: "loss1_c"
  type: "SoftmaxWithLoss"
  bottom: "conv1_cl"
  bottom: "label"
  top: "loss1_c"
  loss_weight: 1
}
layer {
  name: "loss1_p"
  type: "SoftmaxWithLoss"
  bottom: "pool1_cl"
  bottom: "label"
  top: "loss1_p"
  loss_weight: 1
}
layer {
  name: "loss2_c"
  type: "SoftmaxWithLoss"
  bottom: "conv2_cl"
  bottom: "label"
  top: "loss2_c"
  loss_weight: 1
}
layer {
  name: "loss2_p"
  type: "SoftmaxWithLoss"
  bottom: "pool2_cl"
  bottom: "label"
  top: "loss2_p"
  loss_weight: 1
}
layer {
  name: "loss3_c"
  type: "SoftmaxWithLoss"
  bottom: "conv3_cl"
  bottom: "label"
  top: "loss3_c"
  loss_weight: 1
}
layer {
  name: "loss4_c"
  type: "SoftmaxWithLoss"
  bottom: "conv4_cl"
  bottom: "label"
  top: "loss4_c"
  loss_weight: 1
}
layer {
  name: "loss5_c"
  type: "SoftmaxWithLoss"
  bottom: "conv5_cl"
  bottom: "label"
  top: "loss5_c"
  loss_weight: 1
}
layer {
  name: "loss5_p"
  type: "SoftmaxWithLoss"
  bottom: "pool5_cl"
  bottom: "label"
  top: "loss5_p"
  loss_weight: 1
}
# ***************************
# ***** ACCURACY LAYERS *****
# ***************************
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "conv1_cl"
  bottom: "label"
  top: "k1_accuracy1_c"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "pool1_cl"
  bottom: "label"
  top: "k1_accuracy1_p"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "conv2_cl"
  bottom: "label"
  top: "k1_accuracy2_c"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "pool2_cl"
  bottom: "label"
  top: "k1_accuracy2_p"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "conv3_cl"
  bottom: "label"
  top: "k1_accuracy3_c"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "conv4_cl"
  bottom: "label"
  top: "k1_accuracy4_c"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "conv5_cl"
  bottom: "label"
  top: "k1_accuracy5_c"
  include {    phase: TEST  }
}
layer {
  name: "k1_accuracy"
  type: "Accuracy"
  bottom: "pool5_cl"
  bottom: "label"
  top: "k1_accuracy5_p"
  include {    phase: TEST  }
}