train.prototxt

name: "SketchTriplet"
layer {
  name: "input_data"
  type: "Python"
  top: "data_a"
  top: "data_p"
  top: "data_n"
  phase: TRAIN
  python_param {
    module: "triplet_data"
    layer: "TripletDataLayer"
    param_str: "{\'scale\': 1.0, \'classes_per_batch\': 20, \'mean_file\': \'/vol/vssp/ddascratch/Tu/classification/pycaffe/clas3_b/db/image250/image_mean.binaryproto\', \'batch_size\': 200, \'img_source\': \'/vol/vssp/ddascratch/Tu/classification/pycaffe/clas3_b/db/image250\', \'skt_source\': \'/vol/vssp/ddascratch/Tu/classification/pycaffe/clas3_b/db/sketch250_all\', \'shape\': [225, 225], \'rot\': [-5, 5]}"
  }
}
layer {
  name: "conv1_a"
  type: "Convolution"
  bottom: "data_a"
  top: "conv1_a"
  param {
    name: "conv1_a_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv1_a_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    pad: 0
    kernel_size: 15
    stride: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu1_a"
  type: "ReLU"
  bottom: "conv1_a"
  top: "conv1_a"
}
layer {
  name: "pool1_a"
  type: "Pooling"
  bottom: "conv1_a"
  top: "pool1_a"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv2_a"
  type: "Convolution"
  bottom: "pool1_a"
  top: "conv2_a"
  param {
    name: "conv2_a_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv2_a_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu2_a"
  type: "ReLU"
  bottom: "conv2_a"
  top: "conv2_a"
}
layer {
  name: "pool2_a"
  type: "Pooling"
  bottom: "conv2_a"
  top: "pool2_a"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv3_a"
  type: "Convolution"
  bottom: "pool2_a"
  top: "conv3_a"
  param {
    name: "conv3_a_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv3_a_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_a"
  type: "ReLU"
  bottom: "conv3_a"
  top: "conv3_a"
}
layer {
  name: "conv4_a"
  type: "Convolution"
  bottom: "conv3_a"
  top: "conv4_a"
  param {
    name: "conv4_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv4_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_a"
  type: "ReLU"
  bottom: "conv4_a"
  top: "conv4_a"
}
layer {
  name: "conv5_a"
  type: "Convolution"
  bottom: "conv4_a"
  top: "conv5_a"
  param {
    name: "conv5_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv5_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu5_a"
  type: "ReLU"
  bottom: "conv5_a"
  top: "conv5_a"
}
layer {
  name: "pool5_a"
  type: "Pooling"
  bottom: "conv5_a"
  top: "pool5_a"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_a"
  type: "InnerProduct"
  bottom: "pool5_a"
  top: "fc6_a"
  param {
    name: "fc6_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc6_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu6_a"
  type: "ReLU"
  bottom: "fc6_a"
  top: "fc6_a"
}
layer {
  name: "drop6_a"
  type: "Dropout"
  bottom: "fc6_a"
  top: "fc6_a"
  dropout_param {
    dropout_ratio: 0.55
  }
}
layer {
  name: "fc7_a"
  type: "InnerProduct"
  bottom: "fc6_a"
  top: "fc7_a"
  param {
    name: "fc7_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc7_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu7_a"
  type: "ReLU"
  bottom: "fc7_a"
  top: "fc7_a"
}
layer {
  name: "drop7_a"
  type: "Dropout"
  bottom: "fc7_a"
  top: "fc7_a"
  dropout_param {
    dropout_ratio: 0.55
  }
}
layer {
  name: "feat_a"
  type: "InnerProduct"
  bottom: "fc7_a"
  top: "feat_a"
  param {
    name: "fc8_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc8_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 100
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv1_p"
  type: "Convolution"
  bottom: "data_p"
  top: "conv1_p"
  param {
    name: "conv1_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv1_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    pad: 0
    kernel_size: 15
    stride: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu1_p"
  type: "ReLU"
  bottom: "conv1_p"
  top: "conv1_p"
}
layer {
  name: "pool1_p"
  type: "Pooling"
  bottom: "conv1_p"
  top: "pool1_p"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv2_p"
  type: "Convolution"
  bottom: "pool1_p"
  top: "conv2_p"
  param {
    name: "conv2_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv2_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu2_p"
  type: "ReLU"
  bottom: "conv2_p"
  top: "conv2_p"
}
layer {
  name: "pool2_p"
  type: "Pooling"
  bottom: "conv2_p"
  top: "pool2_p"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv3_p"
  type: "Convolution"
  bottom: "pool2_p"
  top: "conv3_p"
  param {
    name: "conv3_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv3_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_p"
  type: "ReLU"
  bottom: "conv3_p"
  top: "conv3_p"
}
layer {
  name: "conv4_p"
  type: "Convolution"
  bottom: "conv3_p"
  top: "conv4_p"
  param {
    name: "conv4_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv4_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_p"
  type: "ReLU"
  bottom: "conv4_p"
  top: "conv4_p"
}
layer {
  name: "conv5_p"
  type: "Convolution"
  bottom: "conv4_p"
  top: "conv5_p"
  param {
    name: "conv5_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv5_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu5_p"
  type: "ReLU"
  bottom: "conv5_p"
  top: "conv5_p"
}
layer {
  name: "pool5_p"
  type: "Pooling"
  bottom: "conv5_p"
  top: "pool5_p"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_p"
  type: "InnerProduct"
  bottom: "pool5_p"
  top: "fc6_p"
  param {
    name: "fc6_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc6_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu6_p"
  type: "ReLU"
  bottom: "fc6_p"
  top: "fc6_p"
}
layer {
  name: "drop6_p"
  type: "Dropout"
  bottom: "fc6_p"
  top: "fc6_p"
  dropout_param {
    dropout_ratio: 0.55
  }
}
layer {
  name: "fc7_p"
  type: "InnerProduct"
  bottom: "fc6_p"
  top: "fc7_p"
  param {
    name: "fc7_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc7_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu7_p"
  type: "ReLU"
  bottom: "fc7_p"
  top: "fc7_p"
}
layer {
  name: "drop7_p"
  type: "Dropout"
  bottom: "fc7_p"
  top: "fc7_p"
  dropout_param {
    dropout_ratio: 0.55
  }
}
layer {
  name: "feat_p"
  type: "InnerProduct"
  bottom: "fc7_p"
  top: "feat_p"
  param {
    name: "fc8_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc8_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 100
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "conv1_n"
  type: "Convolution"
  bottom: "data_n"
  top: "conv1_n"
  param {
    name: "conv1_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv1_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    pad: 0
    kernel_size: 15
    stride: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu1_n"
  type: "ReLU"
  bottom: "conv1_n"
  top: "conv1_n"
}
layer {
  name: "pool1_n"
  type: "Pooling"
  bottom: "conv1_n"
  top: "pool1_n"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv2_n"
  type: "Convolution"
  bottom: "pool1_n"
  top: "conv2_n"
  param {
    name: "conv2_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv2_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu2_n"
  type: "ReLU"
  bottom: "conv2_n"
  top: "conv2_n"
}
layer {
  name: "pool2_n"
  type: "Pooling"
  bottom: "conv2_n"
  top: "pool2_n"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "conv3_n"
  type: "Convolution"
  bottom: "pool2_n"
  top: "conv3_n"
  param {
    name: "conv3_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv3_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu3_n"
  type: "ReLU"
  bottom: "conv3_n"
  top: "conv3_n"
}
layer {
  name: "conv4_n"
  type: "Convolution"
  bottom: "conv3_n"
  top: "conv4_n"
  param {
    name: "conv4_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv4_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu4_n"
  type: "ReLU"
  bottom: "conv4_n"
  top: "conv4_n"
}
layer {
  name: "conv5_n"
  type: "Convolution"
  bottom: "conv4_n"
  top: "conv5_n"
  param {
    name: "conv5_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "conv5_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 0
    kernel_size: 3
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu5_n"
  type: "ReLU"
  bottom: "conv5_n"
  top: "conv5_n"
}
layer {
  name: "pool5_n"
  type: "Pooling"
  bottom: "conv5_n"
  top: "pool5_n"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6_n"
  type: "InnerProduct"
  bottom: "pool5_n"
  top: "fc6_n"
  param {
    name: "fc6_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc6_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu6_n"
  type: "ReLU"
  bottom: "fc6_n"
  top: "fc6_n"
}
layer {
  name: "drop6_n"
  type: "Dropout"
  bottom: "fc6_n"
  top: "fc6_n"
  dropout_param {
    dropout_ratio: 0.55
  }
}
layer {
  name: "fc7_n"
  type: "InnerProduct"
  bottom: "fc6_n"
  top: "fc7_n"
  param {
    name: "fc7_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc7_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 512
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "relu7_n"
  type: "ReLU"
  bottom: "fc7_n"
  top: "fc7_n"
}
layer {
  name: "drop7_n"
  type: "Dropout"
  bottom: "fc7_n"
  top: "fc7_n"
  dropout_param {
    dropout_ratio: 0.55
  }
}
layer {
  name: "feat_n"
  type: "InnerProduct"
  bottom: "fc7_n"
  top: "feat_n"
  param {
    name: "fc8_p_w"
    lr_mult: 1.0
    decay_mult: 1
  }
  param {
    name: "fc8_p_b"
    lr_mult: 2.0
    decay_mult: 0
  }
  inner_product_param {
    num_output: 100
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0.0
    }
  }
}
layer {
  name: "triplet_loss"
  type: "Python"
  bottom: "feat_a"
  bottom: "feat_p"
  bottom: "feat_n"
  top: "loss"
  loss_weight: 1.0
  python_param {
    module: "triplet_loss"
    layer: "TripletLossLayer1"
    param_str: "{\'anchor_scale\': 2.0, \'loss_weight\': 1.0, \'margin\': 1.0}"
  }
}