SfSNet_deploy.prototxt

name : "PS-Net"
#data


layer {
  name: "data"
  type: "Input"
  top: "data"
    input_param { shape: { dim: 1 dim: 3 dim: 128 dim: 128 } }

}

############################ Initial
#C64
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
name : "c1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name : "c1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 7
    stride: 1
    pad: 3
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "bn1"
  type: "BatchNorm"
  bottom: "conv1"
  top: "conv1"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "b1_a"
    lr_mult: 0
  }
  param {
name: "b1_b"
    lr_mult: 0
  }
  param {
name: "b1_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "bn1"
  type: "BatchNorm"
  bottom: "conv1"
  top: "conv1"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "b1_a"
    lr_mult: 0
  }
  param {
name : "b1_b"
    lr_mult: 0
  }
  param {
name: "b1_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

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

#C128
layer {
  name: "conv2"
  type: "Convolution"
  bottom: "conv1"
  top: "conv2"
  param {
name: "c2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "c2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "bn2"
  type: "BatchNorm"
  bottom: "conv2"
  top: "conv2"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "b2_a"
    lr_mult: 0
  }
  param {
name : "b2_b"
    lr_mult: 0
  }
  param {
name: "b2_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "bn2"
  type: "BatchNorm"
  bottom: "conv2"
  top: "conv2"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "b2_a"
    lr_mult: 0
  }
  param {
name : "b2_b"
    lr_mult: 0
  }
  param {
name: "b2_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "relu2"
  type: "ReLU"
  bottom: "conv2"
  top: "conv2"
}

#C128 S2
layer {
  name: "conv3"
  type: "Convolution"
  bottom: "conv2"
  top: "conv3"
  param {
name : "c3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name : "c3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 2
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

###################################################### RESNET for normals



####### RES1


layer {
  name: "nbn1"
  type: "BatchNorm"
  bottom: "conv3"
  top: "nbn1"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb1_a"
    lr_mult: 0
  }
  param {
name: "nb1_b"
    lr_mult: 0
  }
  param {
name: "nb1_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn1"
  type: "BatchNorm"
  bottom: "conv3"
  top: "nbn1"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb1_a"
    lr_mult: 0
  }
  param {
name: "nb1_b"
    lr_mult: 0
  }
  param {
name: "nb1_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu1"
  type: "ReLU"
  bottom: "nbn1"
  top: "nbn1"
}



layer {
  name: "nconv1"
  type: "Convolution"
  bottom: "nbn1"
  top: "nconv1"
  param {
name : "nc1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "nbn1r"
  type: "BatchNorm"
  bottom: "nconv1"
  top: "nconv1"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb1r_a"
    lr_mult: 0
  }
  param {
name: "nb1r_b"
    lr_mult: 0
  }
  param {
name: "nb1r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn1r"
  type: "BatchNorm"
  bottom: "nconv1"
  top: "nconv1"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb1r_a"
    lr_mult: 0
  }
  param {
name: "nb1r_b"
    lr_mult: 0
  }
  param {
name: "nb1r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu1r"
  type: "ReLU"
  bottom: "nconv1"
  top: "nconv1"
}

layer {
  name: "nconv1r"
  type: "Convolution"
  bottom: "nconv1"
  top: "nconv1r"
  param {
name: "nc1r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc1r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "nsum1"
  type: "Eltwise"
  bottom: "nconv1r"
  bottom: "conv3"
  top: "nsum1"
  eltwise_param { operation: SUM }
}


####### RES2



layer {
  name: "nbn2"
  type: "BatchNorm"
  bottom: "nsum1"
  top: "nbn2"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb2_a"
    lr_mult: 0
  }
  param {
name: "nb2_b"
    lr_mult: 0
  }
  param {
name: "nb2_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn2"
  type: "BatchNorm"
  bottom: "nsum1"
  top: "nbn2"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb2_a"
    lr_mult: 0
  }
  param {
name: "nb2_b"
    lr_mult: 0
  }
  param {
name: "nb2_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu2"
  type: "ReLU"
  bottom: "nbn2"
  top: "nbn2"
}


layer {
  name: "nconv2"
  type: "Convolution"
  bottom: "nbn2"
  top: "nconv2"
  param {
name : "nc2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "nbn2r"
  type: "BatchNorm"
  bottom: "nconv2"
  top: "nconv2"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb2r_a"
    lr_mult: 0
  }
  param {
name: "nb2r_b"
    lr_mult: 0
  }
  param {
name: "nb2r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn2r"
  type: "BatchNorm"
  bottom: "nconv2"
  top: "nconv2"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb2r_a"
    lr_mult: 0
  }
  param {
name: "nb2r_b"
    lr_mult: 0
  }
  param {
name: "nb2r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu2r"
  type: "ReLU"
  bottom: "nconv2"
  top: "nconv2"
}

layer {
  name: "nconv2r"
  type: "Convolution"
  bottom: "nconv2"
  top: "nconv2r"
  param {
name : "nc2r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc2r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "nsum2"
  type: "Eltwise"
  bottom: "nconv2r"
  bottom: "nsum1"
  top: "nsum2"
  eltwise_param { operation: SUM }
}

####### RES3



layer {
  name: "nbn3"
  type: "BatchNorm"
  bottom: "nsum2"
  top: "nbn3"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb3_a"
    lr_mult: 0
  }
  param {
name: "nb3_b"
    lr_mult: 0
  }
  param {
name: "nb3_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn3"
  type: "BatchNorm"
  bottom: "nsum2"
  top: "nbn3"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb3_a"
    lr_mult: 0
  }
  param {
name: "nb3_b"
    lr_mult: 0
  }
  param {
name: "nb3_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu3"
  type: "ReLU"
  bottom: "nbn3"
  top: "nbn3"
}


layer {
  name: "nconv3"
  type: "Convolution"
  bottom: "nbn3"
  top: "nconv3"
  param {
name : "nc3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "nbn3r"
  type: "BatchNorm"
  bottom: "nconv3"
  top: "nconv3"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb3r_a"
    lr_mult: 0
  }
  param {
name: "nb3r_b"
    lr_mult: 0
  }
  param {
name: "nb3r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn3r"
  type: "BatchNorm"
  bottom: "nconv3"
  top: "nconv3"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb3r_a"
    lr_mult: 0
  }
  param {
name: "nb3r_b"
    lr_mult: 0
  }
  param {
name: "nb3r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu3r"
  type: "ReLU"
  bottom: "nconv3"
  top: "nconv3"
}

layer {
  name: "nconv3r"
  type: "Convolution"
  bottom: "nconv3"
  top: "nconv3r"
  param {
name : "nc3r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc3r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "nsum3"
  type: "Eltwise"
  bottom: "nconv3r"
  bottom: "nsum2"
  top: "nsum3"
  eltwise_param { operation: SUM }
}


####### RES4



layer {
  name: "nbn4"
  type: "BatchNorm"
  bottom: "nsum3"
  top: "nbn4"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb4_a"
    lr_mult: 0
  }
  param {
name: "nb4_b"
    lr_mult: 0
  }
  param {
name: "nb4_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn4"
  type: "BatchNorm"
  bottom: "nsum3"
  top: "nbn4"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb4_a"
    lr_mult: 0
  }
  param {
name: "nb4_b"
    lr_mult: 0
  }
  param {
name: "nb4_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu4"
  type: "ReLU"
  bottom: "nbn4"
  top: "nbn4"
}


layer {
  name: "nconv4"
  type: "Convolution"
  bottom: "nbn4"
  top: "nconv4"
  param {
name : "nc4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "nbn4r"
  type: "BatchNorm"
  bottom: "nconv4"
  top: "nconv4"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb4r_a"
    lr_mult: 0
  }
  param {
name: "nb4r_b"
    lr_mult: 0
  }
  param {
name: "nb4r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn4r"
  type: "BatchNorm"
  bottom: "nconv4"
  top: "nconv4"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb4r_a"
    lr_mult: 0
  }
  param {
name: "nb4r_b"
    lr_mult: 0
  }
  param {
name: "nb4r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu4r"
  type: "ReLU"
  bottom: "nconv4"
  top: "nconv4"
}

layer {
  name: "nconv4r"
  type: "Convolution"
  bottom: "nconv4"
  top: "nconv4r"
  param {
name : "nc4r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc4r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "nsum4"
  type: "Eltwise"
  bottom: "nconv4r"
  bottom: "nsum3"
  top: "nsum4"
  eltwise_param { operation: SUM }
}

####### RES5



layer {
  name: "nbn5"
  type: "BatchNorm"
  bottom: "nsum4"
  top: "nbn5"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb5_a"
    lr_mult: 0
  }
  param {
name: "nb5_b"
    lr_mult: 0
  }
  param {
name: "nb5_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn5"
  type: "BatchNorm"
  bottom: "nsum4"
  top: "nbn5"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb5_a"
    lr_mult: 0
  }
  param {
name: "nb5_b"
    lr_mult: 0
  }
  param {
name: "nb5_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu5"
  type: "ReLU"
  bottom: "nbn5"
  top: "nbn5"
}


layer {
  name: "nconv5"
  type: "Convolution"
  bottom: "nbn5"
  top: "nconv5"
  param {
name : "nc5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "nbn5r"
  type: "BatchNorm"
  bottom: "nconv5"
  top: "nconv5"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb5r_a"
    lr_mult: 0
  }
  param {
name: "nb5r_b"
    lr_mult: 0
  }
  param {
name: "nb5r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn5r"
  type: "BatchNorm"
  bottom: "nconv5"
  top: "nconv5"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb5r_a"
    lr_mult: 0
  }
  param {
name: "nb5r_b"
    lr_mult: 0
  }
  param {
name: "nb5r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu5r"
  type: "ReLU"
  bottom: "nconv5"
  top: "nconv5"
}

layer {
  name: "nconv5r"
  type: "Convolution"
  bottom: "nconv5"
  top: "nconv5r"
  param {
name : "nc5r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc5r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "nsum5"
  type: "Eltwise"
  bottom: "nconv5r"
  bottom: "nsum4"
  top: "nsum5"
  eltwise_param { operation: SUM }
}

layer {
  name: "nbn6r"
  type: "BatchNorm"
  bottom: "nsum5"
  top: "nsum5"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "nb6r_a"
    lr_mult: 0
  }
  param {
name: "nb6r_b"
    lr_mult: 0
  }
  param {
name: "nb6r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn6r"
  type: "BatchNorm"
  bottom: "nsum5"
  top: "nsum5"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "nb6r_a"
    lr_mult: 0
  }
  param {
name: "nb6r_b"
    lr_mult: 0
  }
  param {
name: "nb6r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu6r"
  type: "ReLU"
  bottom: "nsum5"
  top: "nsum5"
}

#CD128
layer {
  name: "nup6"
  type: "Deconvolution"
  bottom: "nsum5" 
  top: "nup6"
  convolution_param {
    kernel_size: 4
    stride: 2
    num_output: 128
    group: 128
    pad: 1
    weight_filler { 
       type: "bilinear" 
    } 
    bias_term: false
  }
  param { 
    lr_mult: 0 
    decay_mult: 0 
  }
}

layer {
  name: "nconv6"
  type: "Convolution"
  bottom: "nup6"
  top: "nconv6"
  param {
name : "nc6_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc6_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "xavier"
    }
  }
}


layer {
  name: "nbn6"
  type: "BatchNorm"
  bottom: "nconv6"
  top: "nconv6"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name: "nb6_a"
    lr_mult: 0
  }
  param {
name: "nb6_b"
    lr_mult: 0
  }
  param {
name: "nb6_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn6"
  type: "BatchNorm"
  bottom: "nconv6"
  top: "nconv6"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name: "nb6_a"
    lr_mult: 0
  }
  param {
name: "nb6_b"
    lr_mult: 0
  }
  param {
name: "nb6_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu6"
  type: "ReLU"
  bottom: "nconv6"
  top: "nconv6"
}

#CD 64
layer {
  name: "nconv7"
  type: "Convolution"
  bottom: "nconv6"
  top: "nconv7"
  param {
name: "nc7_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "nc7_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "nbn7"
  type: "BatchNorm"
  bottom: "nconv7"
  top: "nconv7"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name: "nb7_a"
    lr_mult: 0
  }
  param {
name: "nb7_b"
    lr_mult: 0
  }
  param {
name: "nb7_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "nbn7"
  type: "BatchNorm"
  bottom: "nconv7"
  top: "nconv7"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name: "nb7_a"
    lr_mult: 0
  }
  param {
name: "nb7_b"
    lr_mult: 0
  }
  param {
name: "nb7_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "nrelu7"
  type: "ReLU"
  bottom: "nconv7"
  top: "nconv7"
}

#C*3
layer {
  name: "Nconv0"
  type: "Convolution"
  bottom: "nconv7"
  top: "Nconv0"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 3
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "xavier"
    }
  }
}


#################################################### Albedo #####################################################





####### RES1


layer {
  name: "abn1"
  type: "BatchNorm"
  bottom: "conv3"
  top: "abn1"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab1_a"
    lr_mult: 0
  }
  param {
name: "ab1_b"
    lr_mult: 0
  }
  param {
name: "ab1_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn1"
  type: "BatchNorm"
  bottom: "conv3"
  top: "abn1"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab1_a"
    lr_mult: 0
  }
  param {
name: "ab1_b"
    lr_mult: 0
  }
  param {
name: "ab1_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu1"
  type: "ReLU"
  bottom: "abn1"
  top: "abn1"
}



layer {
  name: "aconv1"
  type: "Convolution"
  bottom: "abn1"
  top: "aconv1"
  param {
name : "ac1_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac1_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "abn1r"
  type: "BatchNorm"
  bottom: "aconv1"
  top: "aconv1"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab1r_a"
    lr_mult: 0
  }
  param {
name: "ab1r_b"
    lr_mult: 0
  }
  param {
name: "ab1r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn1r"
  type: "BatchNorm"
  bottom: "aconv1"
  top: "aconv1"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab1r_a"
    lr_mult: 0
  }
  param {
name: "ab1r_b"
    lr_mult: 0
  }
  param {
name: "ab1r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu1r"
  type: "ReLU"
  bottom: "aconv1"
  top: "aconv1"
}

layer {
  name: "aconv1r"
  type: "Convolution"
  bottom: "aconv1"
  top: "aconv1r"
  param {
name: "ac1r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac1r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "asum1"
  type: "Eltwise"
  bottom: "aconv1r"
  bottom: "conv3"
  top: "asum1"
  eltwise_param { operation: SUM }
}


####### RES2



layer {
  name: "abn2"
  type: "BatchNorm"
  bottom: "asum1"
  top: "abn2"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab2_a"
    lr_mult: 0
  }
  param {
name: "ab2_b"
    lr_mult: 0
  }
  param {
name: "ab2_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn2"
  type: "BatchNorm"
  bottom: "asum1"
  top: "abn2"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab2_a"
    lr_mult: 0
  }
  param {
name: "ab2_b"
    lr_mult: 0
  }
  param {
name: "ab2_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu2"
  type: "ReLU"
  bottom: "abn2"
  top: "abn2"
}


layer {
  name: "aconv2"
  type: "Convolution"
  bottom: "abn2"
  top: "aconv2"
  param {
name : "ac2_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac2_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "abn2r"
  type: "BatchNorm"
  bottom: "aconv2"
  top: "aconv2"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab2r_a"
    lr_mult: 0
  }
  param {
name: "ab2r_b"
    lr_mult: 0
  }
  param {
name: "ab2r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn2r"
  type: "BatchNorm"
  bottom: "aconv2"
  top: "aconv2"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab2r_a"
    lr_mult: 0
  }
  param {
name: "ab2r_b"
    lr_mult: 0
  }
  param {
name: "ab2r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu2r"
  type: "ReLU"
  bottom: "aconv2"
  top: "aconv2"
}

layer {
  name: "aconv2r"
  type: "Convolution"
  bottom: "aconv2"
  top: "aconv2r"
  param {
name : "ac2r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac2r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "asum2"
  type: "Eltwise"
  bottom: "aconv2r"
  bottom: "asum1"
  top: "asum2"
  eltwise_param { operation: SUM }
}

####### RES3



layer {
  name: "abn3"
  type: "BatchNorm"
  bottom: "asum2"
  top: "abn3"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab3_a"
    lr_mult: 0
  }
  param {
name: "ab3_b"
    lr_mult: 0
  }
  param {
name: "ab3_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn3"
  type: "BatchNorm"
  bottom: "asum2"
  top: "abn3"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab3_a"
    lr_mult: 0
  }
  param {
name: "ab3_b"
    lr_mult: 0
  }
  param {
name: "ab3_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu3"
  type: "ReLU"
  bottom: "abn3"
  top: "abn3"
}


layer {
  name: "aconv3"
  type: "Convolution"
  bottom: "abn3"
  top: "aconv3"
  param {
name : "ac3_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac3_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "abn3r"
  type: "BatchNorm"
  bottom: "aconv3"
  top: "aconv3"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab3r_a"
    lr_mult: 0
  }
  param {
name: "ab3r_b"
    lr_mult: 0
  }
  param {
name: "ab3r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn3r"
  type: "BatchNorm"
  bottom: "aconv3"
  top: "aconv3"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab3r_a"
    lr_mult: 0
  }
  param {
name: "ab3r_b"
    lr_mult: 0
  }
  param {
name: "ab3r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu3r"
  type: "ReLU"
  bottom: "aconv3"
  top: "aconv3"
}

layer {
  name: "aconv3r"
  type: "Convolution"
  bottom: "aconv3"
  top: "aconv3r"
  param {
name : "ac3r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac3r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "asum3"
  type: "Eltwise"
  bottom: "aconv3r"
  bottom: "asum2"
  top: "asum3"
  eltwise_param { operation: SUM }
}


####### RES4



layer {
  name: "abn4"
  type: "BatchNorm"
  bottom: "asum3"
  top: "abn4"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab4_a"
    lr_mult: 0
  }
  param {
name: "ab4_b"
    lr_mult: 0
  }
  param {
name: "ab4_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn4"
  type: "BatchNorm"
  bottom: "asum3"
  top: "abn4"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab4_a"
    lr_mult: 0
  }
  param {
name: "ab4_b"
    lr_mult: 0
  }
  param {
name: "ab4_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu4"
  type: "ReLU"
  bottom: "abn4"
  top: "abn4"
}


layer {
  name: "aconv4"
  type: "Convolution"
  bottom: "abn4"
  top: "aconv4"
  param {
name : "ac4_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac4_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "abn4r"
  type: "BatchNorm"
  bottom: "aconv4"
  top: "aconv4"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab4r_a"
    lr_mult: 0
  }
  param {
name: "ab4r_b"
    lr_mult: 0
  }
  param {
name: "ab4r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn4r"
  type: "BatchNorm"
  bottom: "aconv4"
  top: "aconv4"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab4r_a"
    lr_mult: 0
  }
  param {
name: "ab4r_b"
    lr_mult: 0
  }
  param {
name: "ab4r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu4r"
  type: "ReLU"
  bottom: "aconv4"
  top: "aconv4"
}

layer {
  name: "aconv4r"
  type: "Convolution"
  bottom: "aconv4"
  top: "aconv4r"
  param {
name : "ac4r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac4r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "asum4"
  type: "Eltwise"
  bottom: "aconv4r"
  bottom: "asum3"
  top: "asum4"
  eltwise_param { operation: SUM }
}

####### RES5



layer {
  name: "abn5"
  type: "BatchNorm"
  bottom: "asum4"
  top: "abn5"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab5_a"
    lr_mult: 0
  }
  param {
name: "ab5_b"
    lr_mult: 0
  }
  param {
name: "ab5_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn5"
  type: "BatchNorm"
  bottom: "asum4"
  top: "abn5"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab5_a"
    lr_mult: 0
  }
  param {
name: "ab5_b"
    lr_mult: 0
  }
  param {
name: "ab5_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu5"
  type: "ReLU"
  bottom: "abn5"
  top: "abn5"
}


layer {
  name: "aconv5"
  type: "Convolution"
  bottom: "abn5"
  top: "aconv5"
  param {
name : "ac5_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac5_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}



layer {
  name: "abn5r"
  type: "BatchNorm"
  bottom: "aconv5"
  top: "aconv5"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab5r_a"
    lr_mult: 0
  }
  param {
name: "ab5r_b"
    lr_mult: 0
  }
  param {
name: "ab5r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn5r"
  type: "BatchNorm"
  bottom: "aconv5"
  top: "aconv5"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab5r_a"
    lr_mult: 0
  }
  param {
name: "ab5r_b"
    lr_mult: 0
  }
  param {
name: "ab5r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu5r"
  type: "ReLU"
  bottom: "aconv5"
  top: "aconv5"
}

layer {
  name: "aconv5r"
  type: "Convolution"
  bottom: "aconv5"
  top: "aconv5r"
  param {
name : "ac5r_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac5r_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "asum5"
  type: "Eltwise"
  bottom: "aconv5r"
  bottom: "asum4"
  top: "asum5"
  eltwise_param { operation: SUM }
}
layer {
  name: "abn6r"
  type: "BatchNorm"
  bottom: "asum5"
  top: "asum5"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name : "ab6r_a"
    lr_mult: 0
  }
  param {
name: "ab6r_b"
    lr_mult: 0
  }
  param {
name: "ab6r_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn6r"
  type: "BatchNorm"
  bottom: "asum5"
  top: "asum5"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name : "ab6r_a"
    lr_mult: 0
  }
  param {
name: "ab6r_b"
    lr_mult: 0
  }
  param {
name: "ab6r_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}
layer {
  name: "arelu6r"
  type: "ReLU"
  bottom: "asum5"
  top: "asum5"
}

#CD128
layer {
  name: "aup6"
  type: "Deconvolution"
  bottom: "asum5" 
  top: "aup6"
  convolution_param {
    kernel_size: 4
    stride: 2
    num_output: 128
    group: 128
    pad: 1
    weight_filler { 
       type: "bilinear" 
    } 
    bias_term: false
  }
  param { 
    lr_mult: 0 
    decay_mult: 0 
  }
}

layer {
  name: "aconv6"
  type: "Convolution"
  bottom: "aup6"
  top: "aconv6"
  param {
name : "ac6_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac6_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 128
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "xavier"
    }
  }
}


layer {
  name: "abn6"
  type: "BatchNorm"
  bottom: "aconv6"
  top: "aconv6"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name: "ab6_a"
    lr_mult: 0
  }
  param {
name: "ab6_b"
    lr_mult: 0
  }
  param {
name: "ab6_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn6"
  type: "BatchNorm"
  bottom: "aconv6"
  top: "aconv6"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name: "ab6_a"
    lr_mult: 0
  }
  param {
name: "ab6_b"
    lr_mult: 0
  }
  param {
name: "ab6_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu6"
  type: "ReLU"
  bottom: "aconv6"
  top: "aconv6"
}

#CD 64
layer {
  name: "aconv7"
  type: "Convolution"
  bottom: "aconv6"
  top: "aconv7"
  param {
name: "ac7_w"
    lr_mult: 1
    decay_mult: 1
  }
  param {
name: "ac7_b"
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 64
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "xavier"
    }
  }
}

layer {
  name: "abn7"
  type: "BatchNorm"
  bottom: "aconv7"
  top: "aconv7"
  batch_norm_param {
    use_global_stats: false
  }
  param {
name: "ab7_a"
    lr_mult: 0
  }
  param {
name: "ab7_b"
    lr_mult: 0
  }
  param {
name: "ab7_c"
    lr_mult: 0
  }
  include {
    phase: TRAIN
  }
}
layer {
  name: "abn7"
  type: "BatchNorm"
  bottom: "aconv7"
  top: "aconv7"
  batch_norm_param {
    use_global_stats: true
  }
  param {
name: "ab7_a"
    lr_mult: 0
  }
  param {
name: "ab7_b"
    lr_mult: 0
  }
  param {
name: "ab7_c"
    lr_mult: 0
  }
  include {
    phase: TEST
  }
}

layer {
  name: "arelu7"
  type: "ReLU"
  bottom: "aconv7"
  top: "aconv7"
}

#C*3
layer {
  name: "Aconv0"
  type: "Convolution"
  bottom: "aconv7"
  top: "Aconv0"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 3
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "xavier"
    }
  }
}


################################## Light#############################################

#concat
layer {
name: "lconcat1"
bottom: "nsum5"
bottom: "asum5"
top: "lconcat1"
type: "Concat"
concat_param {
axis: 1
}
}

layer {
name: "lconcat2"
bottom: "lconcat1"
bottom: "conv3"
top: "lconcat2"
type: "Concat"
concat_param {
axis: 1
}
}

#128x1x1 conv
layer {
name: "lconv1"
type: "Convolution"
bottom: "lconcat2"
top: "lconv1"
param {
name : "lc1_w"
lr_mult: 1
decay_mult: 1
}
param {
name: "lc1_b"
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
stride: 1
pad: 0
weight_filler {
type: "xavier"
}
}
}

layer {
name: "lbn1"
type: "BatchNorm"
bottom: "lconv1"
top: "lconv1"
batch_norm_param {
use_global_stats: false
}
param {
name: "lb1_a"
lr_mult: 0
}
param {
name: "lb1_b"
lr_mult: 0
}
param {
name: "lb1_c"
lr_mult: 0
}
include {
phase: TRAIN
}
}
layer {
name: "lbn1"
type: "BatchNorm"
bottom: "lconv1"
top: "lconv1"
batch_norm_param {
use_global_stats: true
}
param {
name: "lb1_a"
lr_mult: 0
}
param {
name: "lb1_b"
lr_mult: 0
}
param {
name: "lb1_c"
lr_mult: 0
}
include {
phase: TEST
}
}

layer {
name: "lrelu1"
type: "ReLU"
bottom: "lconv1"
top: "lconv1"
}

layer {
name: "lpool2r"
type: "Pooling"
bottom: "lconv1"
top: "lpool2r"
pooling_param {
pool: AVE
kernel_size: 64
}
}

layer {
name: "fc_light"
type: "InnerProduct"
bottom: "lpool2r"
top: "fc_light"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 27
weight_filler {
type: "gaussian"
std: 0.005
}
bias_filler {
type: "constant"
value: 1
}
}
}