Large DAU-ConvNet: ImageNet pre-trained model of DAU-ConvNet with AlexNet architecture

readme.md

Information

An DAU-ConvNet with AlexNet architecture with conv3,conv4 and conv5 replaced with DAU-Conv layer. Corresponds to "Large DAU-ConvNet" in CVPR18 paper using larger number of DAUs per channel.

Model is trained on ilsvrc12 and achieves around 57.3% top-1 accuracy.

caffemodel: AlexNet-DAU-ConvNet-large

caffemodel_url: http://box.vicos.si/skokec/caffe-dau-models/alexnet_dau_large.caffemodel

mean_file: use standard ilsvrc 2012 mean file provided by caffe

Caffe compatibility

Compatibile only with https://github.com/skokec/DAU-ConvNet-caffe

alexnet_dau_conv_large.prototxt

name: "AlexNet-DAU-large"
force_backward: true
layer {
  name: "data"
  type: "Input"
  top: "data"
  input_param { shape: { dim: 1 dim: 3 dim: 227 dim: 227 } }
}

layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  convolution_param {
    num_output: 96
    bias_term: false
    pad: 5
    kernel_size: 11
    group: 1
    stride: 4
    weight_filler {
      type: "xavier"
    }
    dilation: 1
  }
}
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: 9.99999974738e-05
    beta: 0.75
  }
}
layer {
  name: "conv2_1"
  type: "DAUConvolution"
  bottom: "norm1"
  top: "conv2_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 5000.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 5000.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 128
    bias_term: true
    number_units: 2
    number_units: 3
    pad: 3
    kernel_size: 7
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 6.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu2_1"
  type: "ReLU"
  bottom: "conv2_1"
  top: "conv2_1"
}
layer {
  name: "conv2_2"
  type: "DAUConvolution"
  bottom: "norm1"
  top: "conv2_2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 5000.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 5000.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 128
    bias_term: true
    number_units: 2
    number_units: 3
    pad: 3
    kernel_size: 7
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 6.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu2_2"
  type: "ReLU"
  bottom: "conv2_2"
  top: "conv2_2"
}
layer {
  name: "conv2"
  type: "Concat"
  bottom: "conv2_1"
  bottom: "conv2_2"
  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: 9.99999974738e-05
    beta: 0.75
  }
}
layer {
  name: "conv3"
  type: "DAUConvolution"
  bottom: "norm2"
  top: "conv3"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 5000.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 5000.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 384
    bias_term: true
    number_units: 2
    number_units: 2
    pad: 4
    kernel_size: 9
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 8.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu3"
  type: "ReLU"
  bottom: "conv3"
  top: "conv3"
}
layer {
  name: "conv4_1"
  type: "DAUConvolution"
  bottom: "conv3"
  top: "conv4_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 192
    bias_term: true
    number_units: 2
    pad: 3
    kernel_size: 7
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 6.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu4_1"
  type: "ReLU"
  bottom: "conv4_1"
  top: "conv4_1"
}
layer {
  name: "conv4_2"
  type: "DAUConvolution"
  bottom: "conv3"
  top: "conv4_2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 192
    bias_term: true
    number_units: 2
    pad: 3
    kernel_size: 7
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 6.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu4_2"
  type: "ReLU"
  bottom: "conv4_2"
  top: "conv4_2"
}
layer {
  name: "conv4"
  type: "Concat"
  bottom: "conv4_1"
  bottom: "conv4_2"
  top: "conv4"
}
layer {
  name: "conv5_1"
  type: "DAUConvolution"
  bottom: "conv4"
  top: "conv5_1"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 128
    bias_term: true
    number_units: 2
    pad: 3
    kernel_size: 7
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 6.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu5_1"
  type: "ReLU"
  bottom: "conv5_1"
  top: "conv5_1"
}
layer {
  name: "conv5_2"
  type: "DAUConvolution"
  bottom: "conv4"
  top: "conv5_2"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 500.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 0.0
    decay_mult: 0.0
  }
  param {
    lr_mult: 2.0
  }
  dau_conv_param {
    num_output: 128
    bias_term: true
    number_units: 2
    pad: 3
    kernel_size: 7
    stride: 1
    weight_filler {
      type: "gaussian"
      std: 0.00999999977648
    }
    bias_filler {
      type: "constant"
      value: 1.0
    }
    mu_filler {
      type: "uniform"
      min: 2.0
      max: 6.0
    }
    sigma_filler {
      type: "constant"
      value: 0.5
    }
    component_border_bound: 0.0
    sigma_lower_bound: 0.300000011921
    engine: CUDNN
  }
}
layer {
  name: "relu5_2"
  type: "ReLU"
  bottom: "conv5_2"
  top: "conv5_2"
}
layer {
  name: "conv5"
  type: "Concat"
  bottom: "conv5_1"
  bottom: "conv5_2"
  top: "conv5"
}
layer {
  name: "pool3"
  type: "Pooling"
  bottom: "conv5"
  top: "pool3"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "fc6"
  type: "InnerProduct"
  bottom: "pool3"
  top: "fc6"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 4096
    bias_term: true
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "relu6"
  type: "ReLU"
  bottom: "fc6"
  top: "fc6"
}
layer {
  name: "fc6_d"
  type: "Dropout"
  bottom: "fc6"
  top: "fc6_d"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc7"
  type: "InnerProduct"
  bottom: "fc6_d"
  top: "fc7"
  param {
    lr_mult: 1.0
    decay_mult: 1.0
  }
  param {
    lr_mult: 2.0
    decay_mult: 0.0
  }
  inner_product_param {
    num_output: 4096
    bias_term: true
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "relu7"
  type: "ReLU"
  bottom: "fc7"
  top: "fc7"
}
layer {
  name: "fc7_d"
  type: "Dropout"
  bottom: "fc7"
  top: "fc7_d"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc8"
  type: "InnerProduct"
  bottom: "fc7_d"
  top: "fc8"
  inner_product_param {
    num_output: 1000
    weight_filler {
      type: "xavier"
    }
  }
}
layer {
  name: "loss"
  type: "Softmax"
  bottom: "fc8"
  top: "loss"
}