h2_train.prototxt

name: "H2_TRAIN"
layer {
  name: "data"
  type: "ImageLabelmapData"
  top: "data"
  top: "label"
  transform_param {
    mirror: false
    mean_value: 104.00699
    mean_value: 116.66877
    mean_value: 122.67892
  }
  image_data_param {
    # SKL
    root_folder: "data/SK-LARGE/"
    source: "data/SK-LARGE/list.txt"
    # PASCAL
    #root_folder: "data/SYM-PASCAL/"
    #source: "data/SYM-PASCAL/train_pair.txt"
    batch_size: 1
    shuffle: true
  }
}
layer { bottom: 'data' top: 'conv1_1' name: 'conv1_1' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 64 pad: 35 kernel_size: 3 } }
layer { bottom: 'conv1_1' top: 'conv1_1' name: 'relu1_1' type: "ReLU" }

layer { bottom: 'conv1_1' top: 'conv1_2' name: 'conv1_2' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 64 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv1_2' top: 'conv1_2' name: 'relu1_2' type: "ReLU" }
layer { name: 'pool1' bottom: 'conv1_2' top: 'pool1' type: "Pooling"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { name: 'conv2_1' bottom: 'pool1' top: 'conv2_1' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 128 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv2_1' top: 'conv2_1' name: 'relu2_1' type: "ReLU" }
layer { bottom: 'conv2_1' top: 'conv2_2' name: 'conv2_2' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 128 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv2_2' top: 'conv2_2' name: 'relu2_2' type: "ReLU" }
layer { bottom: 'conv2_2' top: 'pool2' name: 'pool2' type: "Pooling"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { bottom: 'pool2' top: 'conv3_1' name: 'conv3_1' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv3_1' top: 'conv3_1' name: 'relu3_1' type: "ReLU" }
layer { bottom: 'conv3_1' top: 'conv3_2' name: 'conv3_2' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv3_2' top: 'conv3_2' name: 'relu3_2' type: "ReLU" }
layer { bottom: 'conv3_2' top: 'conv3_3' name: 'conv3_3' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv3_3' top: 'conv3_3' name: 'relu3_3' type: "ReLU" }
layer { bottom: 'conv3_3' top: 'pool3' name: 'pool3' type: "Pooling"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { bottom: 'pool3' top: 'conv4_1' name: 'conv4_1' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv4_1' top: 'conv4_1' name: 'relu4_1' type: "ReLU" }
layer { bottom: 'conv4_1' top: 'conv4_2' name: 'conv4_2' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv4_2' top: 'conv4_2' name: 'relu4_2' type: "ReLU" }
layer { bottom: 'conv4_2' top: 'conv4_3' name: 'conv4_3' type: "Convolution"
  param { lr_mult: 1 decay_mult: 1 } param { lr_mult: 2 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv4_3' top: 'conv4_3' name: 'relu4_3' type: "ReLU" }
layer { bottom: 'conv4_3' top: 'pool4' name: 'pool4' type: "Pooling"
  pooling_param { pool: MAX kernel_size: 2 stride: 2 } }

layer { bottom: 'pool4' top: 'conv5_1' name: 'conv5_1' type: "Convolution"
  param { lr_mult: 100 decay_mult: 1 } param { lr_mult: 200 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv5_1' top: 'conv5_1' name: 'relu5_1' type: "ReLU" }

layer { bottom: 'conv5_1' top: 'conv5_2' name: 'conv5_2' type: "Convolution"
  param { lr_mult: 100 decay_mult: 1 } param { lr_mult: 200 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv5_2' top: 'conv5_2' name: 'relu5_2' type: "ReLU" }

layer { bottom: 'conv5_2' top: 'conv5_3' name: 'conv5_3' type: "Convolution"
  param { lr_mult: 100 decay_mult: 1 } param { lr_mult: 200 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }
layer { bottom: 'conv5_3' top: 'conv5_3' name: 'relu5_3' type: "ReLU" }

## 1top
layer {
  name: "w1_1top" type: "Convolution" bottom: "conv1_1" top: "w1_1top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w1_2top" type: "Convolution" bottom: "conv1_2" top: "w1_2top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 2top
layer {
  name: "w2_1top" type: "Convolution" bottom: "conv2_1" top: "w2_1top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w2_2top" type: "Convolution" bottom: "conv2_2" top: "w2_2top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 2down
layer {
  name: "w2_1down" type: "Convolution" bottom: "conv2_1" top: "w2_1down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w2_2down" type: "Convolution" bottom: "conv2_2" top: "w2_2down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 3top
layer {
  name: "w3_1top" type: "Convolution" bottom: "conv3_1" top: "w3_1top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w3_2top" type: "Convolution" bottom: "conv3_2" top: "w3_2top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w3_3top" type: "Convolution" bottom: "conv3_3" top: "w3_3top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 3down
layer {
  name: "w3_1down" type: "Convolution" bottom: "conv3_1" top: "w3_1down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w3_2down" type: "Convolution" bottom: "conv3_2" top: "w3_2down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w3_3down" type: "Convolution" bottom: "conv3_3" top: "w3_3down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 4top
layer {
  name: "w4_1top" type: "Convolution" bottom: "conv4_1" top: "w4_1top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w4_2top" type: "Convolution" bottom: "conv4_2" top: "w4_2top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w4_3top" type: "Convolution" bottom: "conv4_3" top: "w4_3top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 4down
layer {
  name: "w4_1down" type: "Convolution" bottom: "conv4_1" top: "w4_1down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w4_2down" type: "Convolution" bottom: "conv4_2" top: "w4_2down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w4_3down" type: "Convolution" bottom: "conv4_3" top: "w4_3down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## 5down
layer {
  name: "w5_1down" type: "Convolution" bottom: "conv5_1" top: "w5_1down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w5_2down" type: "Convolution" bottom: "conv5_2" top: "w5_2down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "w5_3down" type: "Convolution" bottom: "conv5_3" top: "w5_3down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
#### upsample
layer {
  name: "upsample2_1" type: "Deconvolution" bottom: "w2_1down" top: "w2_1down_upsample"
  convolution_param{ kernel_size: 4 stride: 2 num_output: 11 }
  param { lr_mult: 0 decay_mult: 0 } param { lr_mult: 0  decay_mult: 0 }
}
layer {
  name: "upsample2_2" type: "Deconvolution" bottom: "w2_2down" top: "w2_2down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample3_1" type: "Deconvolution" bottom: "w3_1down" top: "w3_1down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample3_2" type: "Deconvolution" bottom: "w3_2down" top: "w3_2down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample3_3" type: "Deconvolution" bottom: "w3_3down" top: "w3_3down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample4_1" type: "Deconvolution" bottom: "w4_1down" top: "w4_1down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample4_2" type: "Deconvolution" bottom: "w4_2down" top: "w4_2down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample4_3" type: "Deconvolution" bottom: "w4_3down" top: "w4_3down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample5_1" type: "Deconvolution" bottom: "w5_1down" top: "w5_1down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample5_2" type: "Deconvolution" bottom: "w5_2down" top: "w5_2down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0  decay_mult: 0}
}
layer {
  name: "upsample5_3" type: "Deconvolution" bottom: "w5_3down" top: "w5_3down_upsample"
  convolution_param{kernel_size: 4 stride: 2 num_output: 11}
  param {lr_mult: 0 decay_mult: 0} param {lr_mult: 0 decay_mult: 0}
}
## crop
layer {
  name: "crop2_1" type: "Crop" bottom: "w2_1down_upsample" bottom: "w1_1top" top: "w2_1down_crop"
}
layer {
  name: "crop2_2" type: "Crop" bottom: "w2_2down_upsample" bottom: "w1_1top" top: "w2_2down_crop"
}
layer {
  name: "crop3_1" type: "Crop" bottom: "w3_1down_upsample" bottom: "w2_1top" top: "w3_1down_crop"
}
layer {
  name: "crop3_2" type: "Crop"  bottom: "w3_2down_upsample" bottom: "w2_1top" top: "w3_2down_crop"
}
layer {
  name: "crop3_3" type: "Crop"  bottom: "w3_3down_upsample" bottom: "w2_1top" top: "w3_3down_crop"
}
layer {
  name: "crop4_1" type: "Crop"  bottom: "w4_1down_upsample" bottom: "w3_1top" top: "w4_1down_crop"
}
layer {
  name: "crop4_2" type: "Crop"  bottom: "w4_2down_upsample" bottom: "w3_1top" top: "w4_2down_crop"
}
layer {
  name: "crop4_3" type: "Crop"  bottom: "w4_3down_upsample" bottom: "w3_1top" top: "w4_3down_crop"
}
layer {
  name: "crop5_1" type: "Crop"  bottom: "w5_1down_upsample" bottom: "w4_1top" top: "w5_1down_crop"
}
layer {
  name: "crop5_2" type: "Crop"  bottom: "w5_2down_upsample" bottom: "w4_1top" top: "w5_2down_crop"
}
layer {
  name: "crop5_3" type: "Crop"  bottom: "w5_3down_upsample" bottom: "w4_1top" top: "w5_3down_crop"
}

## Combination hier1
layer {
  name: "h1s1_2" type: "Eltwise" bottom: "w1_1top" bottom: "w1_2top" 
  bottom: "w2_1down_crop" bottom: "w2_2down_crop" top: "h1s1_2"
}
layer {
  name: "h1s2_3" type: "Eltwise" bottom: "w2_1top" bottom: "w2_2top" 
  bottom: "w3_1down_crop" bottom: "w3_2down_crop" bottom: "w3_3down_crop" top: "h1s2_3"
}
layer {
  name: "h1s3_4" type: "Eltwise" bottom: "w3_1top" bottom: "w3_2top" bottom: "w3_3top"
  bottom: "w4_1down_crop" bottom: "w4_2down_crop" bottom: "w4_3down_crop" top: "h1s3_4"
}
layer {
  name: "h1s4_5" type: "Eltwise" bottom: "w4_1top" bottom: "w4_2top" bottom: "w4_3top"
  bottom: "w5_1down_crop" bottom: "w5_2down_crop" bottom: "w5_3down_crop" top: "h1s4_5"
}

## score h1sx_x
layer { name:"score_h1s1_2" type: "Convolution" bottom: "h1s1_2" top: "score_h1s1_2"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 2 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}
layer { name:"score_h1s2_3" type: "Convolution" bottom: "h1s2_3" top: "score_h1s2_3"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 3 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}
layer { name:"score_h1s3_4" type: "Convolution" bottom: "h1s3_4" top: "score_h1s3_4"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 4 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}
layer { name:"score_h1s4_5" type: "Convolution" bottom: "h1s4_5" top: "score_h1s4_5"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 5 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}
# upsample h1sx_x
layer { type: "Deconvolution" name: 'upsample_h1s2_3' bottom: 'score_h1s2_3' top: 'score_h1s2_3_up'
  param { lr_mult: 0 decay_mult: 1 } param { lr_mult: 0 decay_mult: 0}
  convolution_param { kernel_size: 4 stride: 2 num_output: 3 }
}
layer { type: "Deconvolution" name: 'upsample_h1s3_4' bottom: 'score_h1s3_4' top: 'score_h1s3_4_up'
  param { lr_mult: 0 decay_mult: 1 } param { lr_mult: 0 decay_mult: 0}
  convolution_param { kernel_size: 8 stride: 4 num_output: 4 }
}
layer { type: "Deconvolution" name: 'upsample_h1s4_5' bottom: 'score_h1s4_5' top: 'score_h1s4_5_up'
  param { lr_mult: 0 decay_mult: 1 } param { lr_mult: 0 decay_mult: 0}
  convolution_param { kernel_size: 16 stride: 8 num_output: 5 }
}

## crop h1sx_x
layer {
  name: "crop_h1s1_2" type: "Crop" bottom: "score_h1s1_2" bottom: "data" top: "crop_h1s1_2"
}
layer {
  name: "crop_h1s2_3" type: "Crop" bottom: "score_h1s2_3_up" bottom: "data" top: "crop_h1s2_3"
}
layer {
  name: "crop_h1s3_4" type: "Crop" bottom: "score_h1s3_4_up" bottom: "data" top: "crop_h1s3_4"
}
layer {
  name: "crop_h1s4_5" type: "Crop" bottom: "score_h1s4_5_up" bottom: "data" top: "crop_h1s4_5"
}
# h1sx_x loss
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h1s1_2"
  bottom: "crop_h1s1_2" 
  bottom: "label" 
  top:"loss_h1s1_2"  
  loss_param {
    normalize: false
  }
}
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h1s2_3"
  bottom: "crop_h1s2_3" 
  bottom: "label"
  top:"loss_h1s2_3"
  loss_param {
    normalize: false
  }
}
layer {
  type: "SoftmaxWithLoss" 
  name: "loss_h1s3_4"
  bottom: "crop_h1s3_4"
  bottom: "label"
  top:"loss_h1s3_4"
  loss_param {
    normalize: false
  }
}
layer {
  type: "SoftmaxWithLoss" 
  name: "loss_h1s4_5"
  bottom: "crop_h1s4_5"
  bottom: "label"
  top:"loss_h1s4_5"
  loss_param {
    normalize: false
  }
}

## slice
layer{
  type: "Slice" name: "slice_h1s1_2"
  bottom: "crop_h1s1_2"
  top: "slice_h1s1_2_0"
  top: "slice_h1s1_2_1"
  slice_param {
    axis: 1
    slice_point: 1
  }
}
layer{
  type: "Slice" name: "slice_h1s2_3"
  bottom: "crop_h1s2_3"
  top: "slice_h1s2_3_0"
  top: "slice_h1s2_3_1"
  top: "slice_h1s2_3_2"
  slice_param {
    axis: 1
    slice_point: 1
    slice_point: 2
  }
}
layer{
  type: "Slice" name: "slice_h1s3_4"
  bottom: "crop_h1s3_4"
  top: "slice_h1s3_4_0"
  top: "slice_h1s3_4_1"
  top: "slice_h1s3_4_2"
  top: "slice_h1s3_4_3"
  slice_param {
    axis: 1
    slice_point: 1
    slice_point: 2
    slice_point: 3
  }
}
layer{
  type: "Slice" name: "slice_h1s4_5"
  bottom: "crop_h1s4_5"
  top: "slice_h1s4_5_0"
  top: "slice_h1s4_5_1"
  top: "slice_h1s4_5_2"
  top: "slice_h1s4_5_3"
  top: "slice_h1s4_5_4"
  slice_param {
    axis: 1
    slice_point: 1
    slice_point: 2
    slice_point: 3
    slice_point: 4
  }
}
## concat
layer {
  name: "h1_cat0" type: "Concat" 
  bottom: "slice_h1s1_2_0" bottom: "slice_h1s2_3_0" bottom: "slice_h1s3_4_0" bottom: "slice_h1s4_5_0"
  top: "h1_cat0" 
  concat_param { concat_dim: 1} 
}
layer {
  name: "h1_cat1" type: "Concat" 
  bottom: "slice_h1s1_2_1" bottom: "slice_h1s2_3_1" bottom: "slice_h1s3_4_1" bottom: "slice_h1s4_5_1"
  top: "h1_cat1" 
  concat_param { concat_dim: 1} 
}
layer {
  name: "h1_cat2" type: "Concat" 
  bottom: "slice_h1s2_3_2" bottom: "slice_h1s3_4_2" bottom: "slice_h1s4_5_2"
  top: "h1_cat2" 
  concat_param { concat_dim: 1} 
}
layer {
  name: "h1_cat3" type: "Concat" 
  bottom: "slice_h1s3_4_3" bottom: "slice_h1s4_5_3"
  top: "h1_cat3" 
  concat_param { concat_dim: 1} 
}
## score concated
layer { 
  name: 'h1_score0' type: "Convolution" 
  bottom: 'h1_cat0' top: 'h1_score0'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.25 } } 
}
layer { 
  name: 'h1_score1' type: "Convolution" 
  bottom: 'h1_cat1' top: 'h1_score1'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.25 } } 
}
layer { 
  name: 'h1_score2' type: "Convolution" 
  bottom: 'h1_cat2' top: 'h1_score2'
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.333 } } 
}
layer { 
  name: 'h1_score3' type: "Convolution" 
  bottom: 'h1_cat3' top: 'h1_score3'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.5 } } 
}
layer { 
  name: 'h1_score4' type: "Convolution" 
  bottom: 'slice_h1s4_5_4' top: 'h1_score4'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 1 } } 
}
## concat for fuse
layer {
  name: "h1_fuse" type: "Concat" 
  bottom: "h1_score0" bottom: "h1_score1" bottom: "h1_score2" bottom: "h1_score3" bottom: "h1_score4"
  top: "h1_fuse" 
  concat_param { concat_dim: 1} 
}
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h1_fuse"
  bottom: "h1_fuse" 
  bottom: "label"
  top:"loss_h1_fuse"  
  loss_param {
    normalize: false
  }
}
###################### H2
layer {
  name: "wh1s1_2top" type: "Convolution" bottom: "h1s1_2" top: "wh1s1_2top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## w h1s2_3
layer {
  name: "wh1s2_3down" type: "Convolution" bottom: "h1s2_3" top: "wh1s2_3down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "wh1s2_3top" type: "Convolution" bottom: "h1s2_3" top: "wh1s2_3top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## w h1s3_4
layer {
  name: "wh1s3_4down" type: "Convolution" bottom: "h1s3_4" top: "wh1s3_4down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
layer {
  name: "wh1s3_4top" type: "Convolution" bottom: "h1s3_4" top: "wh1s3_4top"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## w h1s4_5
layer {
  name: "wh1s4_5down" type: "Convolution" bottom: "h1s4_5" top: "wh1s4_5down"
  param{lr_mult: 0.1 decay_mult: 1} param{lr_mult: 0.2 decay_mult: 0}
  convolution_param{kernel_size: 1 num_output: 11}
}
## upsample
layer {
  name: "upsample_wh1s2_3" type: "Deconvolution" bottom: "wh1s2_3down" top: "upsample_wh1s2_3"
  convolution_param{ kernel_size: 4 stride: 2 num_output: 11 }
  param { lr_mult: 0 decay_mult: 0 } param { lr_mult: 0  decay_mult: 0 }
}
layer {
  name: "upsample_wh1s3_4" type: "Deconvolution" bottom: "wh1s3_4down" top: "upsample_wh1s3_4"
  convolution_param{ kernel_size: 4 stride: 2 num_output: 11 }
  param { lr_mult: 0 decay_mult: 0 } param { lr_mult: 0  decay_mult: 0 }
}
layer {
  name: "upsample_wh1s4_5" type: "Deconvolution" bottom: "wh1s4_5down" top: "upsample_wh1s4_5"
  convolution_param{ kernel_size: 4 stride: 2 num_output: 11 }
  param { lr_mult: 0 decay_mult: 0 } param { lr_mult: 0  decay_mult: 0 }
}
## crop
layer {
  name: "crop_wh1s2_3" type: "Crop" bottom: "upsample_wh1s2_3" bottom: "wh1s1_2top" top: "crop_wh1s2_3"
}
layer {
  name: "crop_wh1s3_4" type: "Crop" bottom: "upsample_wh1s3_4" bottom: "wh1s2_3top" top: "crop_wh1s3_4"
}
layer {
  name: "crop_wh1s4_5" type: "Crop" bottom: "upsample_wh1s4_5" bottom: "wh1s3_4top" top: "crop_wh1s4_5"
}
## Combination hier12
layer {
  name: "h2s1_2_3" type: "Eltwise" bottom: "wh1s1_2top" bottom: "crop_wh1s2_3" top: "h2s1_2_3"
}
layer {
  name: "h2s2_3_4" type: "Eltwise" bottom: "wh1s2_3top" bottom: "crop_wh1s3_4" top: "h2s2_3_4"
}
layer {
  name: "h2s3_4_5" type: "Eltwise" bottom: "wh1s3_4top" bottom: "crop_wh1s4_5" top: "h2s3_4_5"
}
## score h2sx_x_x
layer { name:"score_h2s1_2_3" type: "Convolution" bottom: "h2s1_2_3" top: "score_h2s1_2_3"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 3 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}
layer { name:"score_h2s2_3_4" type: "Convolution" bottom: "h2s2_3_4" top: "score_h2s2_3_4"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 4 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}
layer { name:"score_h2s3_4_5" type: "Convolution" bottom: "h2s3_4_5" top: "score_h2s3_4_5"
  param { lr_mult: 0.01 decay_mult: 1 } param { lr_mult: 0.02 decay_mult: 0 }
  convolution_param { num_output: 5 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } } 
}

# upsample h2sx_x
layer { type: "Deconvolution" name: 'upscore_h2s2_3_4' bottom: 'score_h2s2_3_4' top: 'upscore_h2s2_3_4'
  param { lr_mult: 0 decay_mult: 1 } param { lr_mult: 0 decay_mult: 0}
  convolution_param { kernel_size: 4 stride: 2 num_output: 4 }
}
layer { type: "Deconvolution" name: 'upscore_h2s3_4_5' bottom: 'score_h2s3_4_5' top: 'upscore_h2s3_4_5'
  param { lr_mult: 0 decay_mult: 1 } param { lr_mult: 0 decay_mult: 0}
  convolution_param { kernel_size: 8 stride: 4 num_output: 5 }
}

## crop h2sx_x_x
layer {
  name: "crop_h2s1_2_3" type: "Crop" bottom: "score_h2s1_2_3" bottom: "data" top: "crop_h2s1_2_3"
}
layer {
  name: "crop_h2s2_3_4" type: "Crop" bottom: "upscore_h2s2_3_4" bottom: "data" top: "crop_h2s2_3_4"
}
layer {
  name: "crop_h2s3_4_5" type: "Crop" bottom: "upscore_h2s3_4_5" bottom: "data" top: "crop_h2s3_4_5"
}
# h2sx_x_x loss
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h2s1_2_3"
  bottom: "crop_h2s1_2_3" 
  bottom: "label" 
  top:"loss_h2s1_2_3"  
  loss_param {
    normalize: false
  }
}
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h2s2_3_4"
  bottom: "crop_h2s2_3_4" 
  bottom: "label" 
  top:"loss_h2s2_3_4"  
  loss_param {
    normalize: false
  }
}
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h2s3_4_5"
  bottom: "crop_h2s3_4_5" 
  bottom: "label" 
  top:"loss_h2s3_4_5"  
  loss_param {
    normalize: false
  }
}
## slice h2sx_x_x
layer{
  type: "Slice" name: "slice_h2s1_2_3"
  bottom: "crop_h2s1_2_3"
  top: "slice_h2s1_2_3_0"
  top: "slice_h2s1_2_3_1"
  top: "slice_h2s1_2_3_2"
  slice_param {
    axis: 1
    slice_point: 1
    slice_point: 2
  }
}
layer{
  type: "Slice" name: "slice_h2s2_3_4"
  bottom: "crop_h2s2_3_4"
  top: "slice_h2s2_3_4_0"
  top: "slice_h2s2_3_4_1"
  top: "slice_h2s2_3_4_2"
  top: "slice_h2s2_3_4_3"
  slice_param {
    axis: 1
    slice_point: 1
    slice_point: 2
    slice_point: 3
  }
}
layer{
  type: "Slice" name: "slice_h2s3_4_5"
  bottom: "crop_h2s3_4_5"
  top: "slice_h2s3_4_5_0"
  top: "slice_h2s3_4_5_1"
  top: "slice_h2s3_4_5_2"
  top: "slice_h2s3_4_5_3"
  top: "slice_h2s3_4_5_4"
  slice_param {
    axis: 1
    slice_point: 1
    slice_point: 2
    slice_point: 3
    slice_point: 4
  }
}
## h2 scale specific concat
layer {
  name: "h2_cat0" type: "Concat" 
  bottom: "slice_h2s1_2_3_0" bottom: "slice_h2s2_3_4_0" bottom: "slice_h2s3_4_5_0" top: "h2_cat0" 
  concat_param { concat_dim: 1} 
}
layer {
  name: "h2_cat1" type: "Concat" 
  bottom: "slice_h2s1_2_3_1" bottom: "slice_h2s2_3_4_1" bottom: "slice_h2s3_4_5_1" top: "h2_cat1" 
  concat_param { concat_dim: 1} 
}
layer {
  name: "h2_cat2" type: "Concat" 
  bottom: "slice_h2s1_2_3_2" bottom: "slice_h2s2_3_4_2" bottom: "slice_h2s3_4_5_2" top: "h2_cat2" 
  concat_param { concat_dim: 1} 
}
layer {
  name: "h2_cat3" type: "Concat" 
  bottom: "slice_h2s2_3_4_3" bottom: "slice_h2s3_4_5_3" top: "h2_cat3" 
  concat_param { concat_dim: 1} 
}
## H2 score scale-specific  concat
layer { 
  name: 'h2_score0' type: "Convolution" 
  bottom: 'h2_cat0' top: 'h2_score0'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.333 } } 
}
layer { 
  name: 'h2_score1' type: "Convolution" 
  bottom: 'h2_cat1' top: 'h2_score1'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.333 } } 
}
layer { 
  name: 'h2_score2' type: "Convolution"
  bottom: 'h2_cat2' top: 'h2_score2'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.333 } } 
}
layer { 
  name: 'h2_score3' type: "Convolution"
  bottom: 'h2_cat3' top: 'h2_score3'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.5 } } 
}
layer { 
  name: 'h2_score4' type: "Convolution"
  bottom: 'slice_h2s3_4_5_4' top: 'h2_score4'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 1 } } 
}
## concat for h2_fuse
layer {
  name: "h2_fuse" type: "Concat" 
  bottom: "h2_score0" bottom: "h2_score1" bottom: "h2_score2" bottom: "h2_score3" bottom: "h2_score4"
  top: "h2_fuse" 
  concat_param { concat_dim: 1} 
}
## loss h2 fuse 
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h2_fuse"
  bottom: "h2_fuse" 
  bottom: "label"
  top:"loss_h2_fuse"  
  loss_param {
    normalize: false
  }
}
## concat H1 and H2
layer {
  name: "h1h2_cat0" type: "Concat" 
  bottom: "slice_h2s1_2_3_0" bottom: "slice_h2s2_3_4_0" bottom: "slice_h2s3_4_5_0" 
  bottom: "slice_h1s1_2_0" bottom: "slice_h1s2_3_0" bottom: "slice_h1s3_4_0" bottom: "slice_h1s4_5_0"
  top: "h1h2_cat0" concat_param { concat_dim: 1} 
}
layer {
  name: "h1h2_cat1" type: "Concat" 
  bottom: "slice_h2s1_2_3_1" bottom: "slice_h2s2_3_4_1" bottom: "slice_h2s3_4_5_1" 
  bottom: "slice_h1s1_2_1" bottom: "slice_h1s2_3_1" bottom: "slice_h1s3_4_1" bottom: "slice_h1s4_5_1"
  top: "h1h2_cat1" concat_param { concat_dim: 1} 
}
layer {
  name: "h1h2_cat2" type: "Concat" 
  bottom: "slice_h2s1_2_3_2" bottom: "slice_h2s2_3_4_2" bottom: "slice_h2s3_4_5_2" 
  bottom: "slice_h1s2_3_2" bottom: "slice_h1s3_4_2" bottom: "slice_h1s4_5_2"
  top: "h1h2_cat2" concat_param { concat_dim: 1} 
}
layer {
  name: "h1h2_cat3" type: "Concat" 
  bottom: "slice_h2s2_3_4_3" bottom: "slice_h2s3_4_5_3" 
  bottom: "slice_h1s3_4_3" bottom: "slice_h1s4_5_3" 
  top: "h1h2_cat3" concat_param { concat_dim: 1} 
}
layer {
  name: "h1h2_cat4" type: "Concat"
  bottom: "slice_h1s4_5_4" bottom: "slice_h2s3_4_5_4"
  top: "h1h2_cat4" concat_param { concat_dim: 1}
}
## H1H2 score scale-specific  concat
layer { 
  name: 'h1h2_score0' type: "Convolution" 
  bottom: 'h1h2_cat0' top: 'h1h2_score0'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.142857 } } 
}
layer { 
  name: 'h1h2_score1' type: "Convolution" 
  bottom: 'h1h2_cat1' top: 'h1h2_score1'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.142857 } } 
}
layer { 
  name: 'h1h2_score2' type: "Convolution"
  bottom: 'h1h2_cat2' top: 'h1h2_score2'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.16667 } } 
}
layer { 
  name: 'h1h2_score3' type: "Convolution"
  bottom: 'h1h2_cat3' top: 'h1h2_score3'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.25 } } 
}
layer { 
  name: 'h1h2_score4' type: "Convolution"
  bottom: 'h1h2_cat4' top: 'h1h2_score4'
  param { lr_mult: 0.05 decay_mult: 1 } param { lr_mult: 0.002 decay_mult: 0}
  convolution_param { engine: CAFFE num_output: 1 kernel_size: 1  weight_filler {type: "constant" value: 0.5 } } 
}
## Concat for H1H2 fuse
layer {
  name: "h2_fuse" type: "Concat" 
  bottom: "h1h2_score0" bottom: "h1h2_score1" bottom: "h1h2_score2" bottom: "h1h2_score3" bottom: "h1h2_score4"
  top: "h1h2_fuse" 
  concat_param { concat_dim: 1} 
}
## loss h1h2 fuse 
layer { 
  type: "SoftmaxWithLoss" 
  name: "loss_h1h2_fuse"
  bottom: "h1h2_fuse" 
  bottom: "label"
  top:"loss_h1h2_fuse"
  loss_param {
    normalize: false
  }
}