HCRF_s1_Coarse_SR.prototxt

name: "STSR"
layer {
  name: "data"
  type: "HDF5Data"
  top: "data"
  top: "label"
  hdf5_data_param {
    source: "examples/DecisionTree_CNN/extra/train.txt"
    batch_size: 8
    shuffle: true
  }
  include: { phase: TRAIN }
}

layer {
  name: "data"
  type: "HDF5Data"
  top: "data"
  top: "label"
  hdf5_data_param {
    source: "examples/DecisionTree_CNN/extra/test.txt"
    batch_size: 1
  }
  include: { phase: TEST }
}

######################### s1 bp0 #########################
layer {
  name: "data_upscale"
  type: "Deconvolution"
  bottom: "data"
  top: "data_upscale"
  param {
    lr_mult: 0
    decay_mult: 0
  }
  convolution_param {
    num_output: 1
    kernel_size: 12
    stride: 8
    pad: 2
    weight_filler {
      type: "bilinear"
    }
    bias_term: false
  }
}

layer {
  name: "conv0_SR_s1_bp0"
  type: "Convolution"
  bottom: "data_upscale"
  top: "conv0_SR_s1_bp0"
  param {
    lr_mult: 1
    name: "conv0_bp0_w"
  }
  param {
    lr_mult: 1
    name: "conv0_bp0_b"
  }
  convolution_param {
    num_output: 256
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.02946
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu0_SR_s1_bp0"
  type: "PReLU"
  bottom: "conv0_SR_s1_bp0"
  top: "conv0_SR_s1_bp0"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "output_SR_s1_bp0"
  type: "Convolution"
  bottom: "conv0_SR_s1_bp0"
  top: "output_SR_s1_bp0"
  param {
    lr_mult: 1
    name: "output_bp0_w"
  }
  param {
    lr_mult: 1
    name: "output_bp0_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp0"
  type: "PReLU"
  bottom: "output_SR_s1_bp0"
  top: "output_SR_s1_bp0"
  prelu_param {
    channel_shared: 1
  }
}

######################### s1 bp1 #########################
layer {
  name: "conv1_SR_s1_bp1"
  type: "Convolution"
  bottom: "output_SR_s1_bp0"
  top: "conv1_SR_s1_bp1"
  param {
    lr_mult: 1
    name: "conv1_s1_bp1_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s1_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp1"
  type: "PReLU"
  bottom: "conv1_SR_s1_bp1"
  top: "conv1_SR_s1_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s1_bp1"
  type: "Deconvolution"
  bottom: "conv1_SR_s1_bp1"
  top: "conv2_SR_s1_bp1"
  param {
    lr_mult: 1
    name: "conv2_s1_bp1_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s1_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s1_bp1"
  type: "PReLU"
  bottom: "conv2_SR_s1_bp1"
  top: "conv2_SR_s1_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s1_bp0"
  type: "Convolution"
  bottom: "output_SR_s1_bp0"
  top: "local_output_SR_s1_bp0"
  param {
    lr_mult: 1
    name: "local_output_s1_bp0_w"
  }
  param {
    lr_mult: 1
    name: "local_output_s1_bp0_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s1_bp1"
    type: "Eltwise"
    bottom: "local_output_SR_s1_bp0"
    bottom: "conv2_SR_s1_bp1"
    top: "eltwise1_SR_s1_bp1"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s1_bp1"
  type: "Convolution"
  bottom: "eltwise1_SR_s1_bp1"
  top: "conv3_SR_s1_bp1"
  param {
    lr_mult: 1
    name: "conv3_s1_bp1_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s1_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s1_bp1"
  type: "PReLU"
  bottom: "conv3_SR_s1_bp1"
  top: "conv3_SR_s1_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s1_bp1"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp1"
  top: "local_conv1_SR_s1_bp1"
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp1_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s1_bp1"
    type: "Eltwise"
    bottom: "local_conv1_SR_s1_bp1"
    bottom: "conv3_SR_s1_bp1"
    top: "output_SR_s1_bp1"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s1_bp1"
  type: "PReLU"
  bottom: "output_SR_s1_bp1"
  top: "output_SR_s1_bp1"
  prelu_param {
    channel_shared: 1
  }
}

######################### s1 bp2 #########################
layer {
  name: "conv1_SR_s1_bp2"
  type: "Convolution"
  bottom: "output_SR_s1_bp1"
  top: "conv1_SR_s1_bp2"
  param {
    lr_mult: 1
    name: "conv1_s1_bp2_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s1_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp2"
  type: "PReLU"
  bottom: "conv1_SR_s1_bp2"
  top: "conv1_SR_s1_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s1_bp2"
  type: "Deconvolution"
  bottom: "conv1_SR_s1_bp2"
  top: "conv2_SR_s1_bp2"
  param {
    lr_mult: 1
    name: "conv2_s1_bp2_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s1_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s1_bp2"
  type: "PReLU"
  bottom: "conv2_SR_s1_bp2"
  top: "conv2_SR_s1_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s1_bp1"
  type: "Convolution"
  bottom: "output_SR_s1_bp1"
  top: "local_output_SR_s1_bp1"
  param {
    lr_mult: 1
    name: "local_output_s1_bp1_w"
  }
  param {
    lr_mult: 1
    name: "local_output_s1_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s1_bp2"
    type: "Eltwise"
    bottom: "local_output_SR_s1_bp1"
    bottom: "conv2_SR_s1_bp2"
    top: "eltwise1_SR_s1_bp2"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s1_bp2"
  type: "Convolution"
  bottom: "eltwise1_SR_s1_bp2"
  top: "conv3_SR_s1_bp2"
  param {
    lr_mult: 1
    name: "conv3_s1_bp2_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s1_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s1_bp2"
  type: "PReLU"
  bottom: "conv3_SR_s1_bp2"
  top: "conv3_SR_s1_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s1_bp2"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp2"
  top: "local_conv1_SR_s1_bp2"
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp2_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s1_bp2"
    type: "Eltwise"
    bottom: "local_conv1_SR_s1_bp2"
    bottom: "conv3_SR_s1_bp2"
    top: "output_SR_s1_bp2"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s1_bp2"
  type: "PReLU"
  bottom: "output_SR_s1_bp2"
  top: "output_SR_s1_bp2"
  prelu_param {
    channel_shared: 1
  }
}

######################### s1 bp3 #########################

layer {
  name: "conv1_SR_s1_bp3"
  type: "Convolution"
  bottom: "output_SR_s1_bp2"
  top: "conv1_SR_s1_bp3"
  param {
    lr_mult: 1
    name: "conv1_s1_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s1_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp3"
  type: "PReLU"
  bottom: "conv1_SR_s1_bp3"
  top: "conv1_SR_s1_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s1_bp3"
  type: "Deconvolution"
  bottom: "conv1_SR_s1_bp3"
  top: "conv2_SR_s1_bp3"
  param {
    lr_mult: 1
    name: "conv2_s1_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s1_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s1_bp3"
  type: "PReLU"
  bottom: "conv2_SR_s1_bp3"
  top: "conv2_SR_s1_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s1_bp2"
  type: "Convolution"
  bottom: "output_SR_s1_bp2"
  top: "local_output_SR_s1_bp2"
  param {
    lr_mult: 1
    name: "local_output_s1_bp2_w"
  }
  param {
    lr_mult: 1
    name: "local_output_s1_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s1_bp3"
    type: "Eltwise"
    bottom: "local_output_SR_s1_bp2"
    bottom: "conv2_SR_s1_bp3"
    top: "eltwise1_SR_s1_bp3"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s1_bp3"
  type: "Convolution"
  bottom: "eltwise1_SR_s1_bp3"
  top: "conv3_SR_s1_bp3"
  param {
    lr_mult: 1
    name: "conv3_s1_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s1_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s1_bp3"
  type: "PReLU"
  bottom: "conv3_SR_s1_bp3"
  top: "conv3_SR_s1_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s1_bp3"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp3"
  top: "local_conv1_SR_s1_bp3"
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s1_bp3"
    type: "Eltwise"
    bottom: "local_conv1_SR_s1_bp3"
    bottom: "conv3_SR_s1_bp3"
    top: "output_SR_s1_bp3"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s1_bp3"
  type: "PReLU"
  bottom: "output_SR_s1_bp3"
  top: "output_SR_s1_bp3"
  prelu_param {
    channel_shared: 1
  }
}

######################### s1 bp3 data duplicate #########################
layer {
  name: "data_duplicate_s1"
  type: "Convolution"
  bottom: "data"
  top: "data_duplicate_s1"
  param {
    lr_mult: 1
    name: "data_duplicate_s1_w"
  }
  param {
    lr_mult: 1
    name: "data_duplicate_s1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      value: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "residual_SR_s1_bp3"
    type: "Eltwise"
    bottom: "output_SR_s1_bp3"
    bottom: "data_duplicate_s1"
    top: "residual_SR_s1_bp3"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

######################### SR_s1_bp4 #########################
layer {
  name: "conv1_SR_s1_bp4"
  type: "Deconvolution"
  bottom: "residual_SR_s1_bp3"
  top: "conv1_SR_s1_bp4"
  param {
    lr_mult: 1
    name: "conv1_s1_bp4_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s1_bp4_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp4"
  type: "PReLU"
  bottom: "conv1_SR_s1_bp4"
  top: "conv1_SR_s1_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s1_bp4"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp4"
  top: "conv2_SR_s1_bp4"
  param {
    lr_mult: 1
    name: "conv2_s1_bp4_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s1_bp4_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s1_bp4"
  type: "PReLU"
  bottom: "conv2_SR_s1_bp4"
  top: "conv2_SR_s1_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s1_bp3"
  type: "Convolution"
  bottom: "residual_SR_s1_bp3"
  top: "local_output_SR_s1_bp3"
  param {
    lr_mult: 1
    name: "local_output_s1_bp3_w"
  }
  param {
    lr_mult: 1
    name: "local_output_s1_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s1_bp4"
    type: "Eltwise"
    bottom: "local_output_SR_s1_bp3"
    bottom: "conv2_SR_s1_bp4"
    top: "eltwise1_SR_s1_bp4"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s1_bp4"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s1_bp4"
  top: "conv3_SR_s1_bp4"
  param {
    lr_mult: 1
    name: "conv3_s1_bp4_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s1_bp4_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s1_bp4"
  type: "PReLU"
  bottom: "conv3_SR_s1_bp4"
  top: "conv3_SR_s1_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s1_bp4"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp4"
  top: "local_conv1_SR_s1_bp4"
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp4_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp4_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s1_bp4"
    type: "Eltwise"
    bottom: "local_conv1_SR_s1_bp4"
    bottom: "conv3_SR_s1_bp4"
    top: "output_SR_s1_bp4"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s1_bp4"
  type: "PReLU"
  bottom: "output_SR_s1_bp4"
  top: "output_SR_s1_bp4"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s1_bp1 & SR_s1_bp4 #########################
layer {
  name: "global_output_SR_s1_bp2"
  type: "Convolution"
  bottom: "output_SR_s1_bp2"
  top: "global_output_SR_s1_bp2"
  param {
    lr_mult: 1
    name: "global_output_s1_bp2_w"
  }
  param {
    lr_mult: 1
    name: "global_output_s1_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "concat_SR_s1_bp2_SR_s1_bp4"
  type: "Concat"
  bottom: "global_output_SR_s1_bp2"
  bottom: "output_SR_s1_bp4"
  top: "concat_SR_s1_bp2_SR_s1_bp4"
}

######################### SR_s1_bp5 #########################
layer {
  name: "conv1_SR_s1_bp5"
  type: "Deconvolution"
  bottom: "concat_SR_s1_bp2_SR_s1_bp4"
  top: "conv1_SR_s1_bp5"
  param {
    lr_mult: 1
    name: "conv1_s1_bp5_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s1_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp5"
  type: "PReLU"
  bottom: "conv1_SR_s1_bp5"
  top: "conv1_SR_s1_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s1_bp5"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp5"
  top: "conv2_SR_s1_bp5"
  param {
    lr_mult: 1
    name: "conv2_s1_bp5_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s1_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s1_bp5"
  type: "PReLU"
  bottom: "conv2_SR_s1_bp5"
  top: "conv2_SR_s1_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s1_bp4"
  type: "Convolution"
  bottom: "concat_SR_s1_bp2_SR_s1_bp4"
  top: "local_output_SR_s1_bp4"
  param {
    lr_mult: 1
    name: "local_output_SR_s1_bp4_w"
  }
  param {
    lr_mult: 1
    name: "local_output_SR_s1_bp4_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s1_bp5"
    type: "Eltwise"
    bottom: "local_output_SR_s1_bp4"
    bottom: "conv2_SR_s1_bp5"
    top: "eltwise1_SR_s1_bp5"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s1_bp5"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s1_bp5"
  top: "conv3_SR_s1_bp5"
  param {
    lr_mult: 1
    name: "conv3_s1_bp5_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s1_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s1_bp5"
  type: "PReLU"
  bottom: "conv3_SR_s1_bp5"
  top: "conv3_SR_s1_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s1_bp5"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp5"
  top: "local_conv1_SR_s1_bp5"
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp5_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s1_bp5"
    type: "Eltwise"
    bottom: "local_conv1_SR_s1_bp5"
    bottom: "conv3_SR_s1_bp5"
    top: "output_SR_s1_bp5"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s1_bp5"
  type: "PReLU"
  bottom: "output_SR_s1_bp5"
  top: "output_SR_s1_bp5"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s1_bp1 & SR_s1_bp5 #########################
layer {
  name: "global_output_SR_s1_bp1"
  type: "Convolution"
  bottom: "output_SR_s1_bp1"
  top: "global_output_SR_s1_bp1"
  param {
    lr_mult: 1
    name: "global_output_s1_bp1_w"
  }
  param {
    lr_mult: 1
    name: "global_output_s1_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "concat_SR_s1_bp1_SR_s1_bp5"
  type: "Concat"
  bottom: "global_output_SR_s1_bp1"
  bottom: "output_SR_s1_bp5"
  top: "concat_SR_s1_bp1_SR_s1_bp5"
}

######################### SR_s1_bp6 #########################
layer {
  name: "conv1_SR_s1_bp6"
  type: "Deconvolution"
  bottom: "concat_SR_s1_bp1_SR_s1_bp5"
  top: "conv1_SR_s1_bp6"
  param {
    lr_mult: 1
    name: "conv1_s1_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s1_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s1_bp6"
  type: "PReLU"
  bottom: "conv1_SR_s1_bp6"
  top: "conv1_SR_s1_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s1_bp6"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp6"
  top: "conv2_SR_s1_bp6"
  param {
    lr_mult: 1
    name: "conv2_s1_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s1_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s1_bp6"
  type: "PReLU"
  bottom: "conv2_SR_s1_bp6"
  top: "conv2_SR_s1_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s1_bp5"
  type: "Convolution"
  bottom: "concat_SR_s1_bp1_SR_s1_bp5"
  top: "local_output_SR_s1_bp5"
  param {
    lr_mult: 1
    name: "local_output_SR_s1_bp5_w"
  }
  param {
    lr_mult: 1
    name: "local_output_SR_s1_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s1_bp6"
    type: "Eltwise"
    bottom: "local_output_SR_s1_bp5"
    bottom: "conv2_SR_s1_bp6"
    top: "eltwise1_SR_s1_bp6"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s1_bp6"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s1_bp6"
  top: "conv3_SR_s1_bp6"
  param {
    lr_mult: 1
    name: "conv3_s1_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s1_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s1_bp6"
  type: "PReLU"
  bottom: "conv3_SR_s1_bp6"
  top: "conv3_SR_s1_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s1_bp6"
  type: "Convolution"
  bottom: "conv1_SR_s1_bp6"
  top: "local_conv1_SR_s1_bp6"
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp6_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s1_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s1_bp6"
    type: "Eltwise"
    bottom: "local_conv1_SR_s1_bp6"
    bottom: "conv3_SR_s1_bp6"
    top: "output_SR_s1_bp6"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s1_bp6"
  type: "PReLU"
  bottom: "output_SR_s1_bp6"
  top: "output_SR_s1_bp6"
  prelu_param {
    channel_shared: 1
  }
}

######################### concate SR_s1 all #########################
layer {
  name: "concat_SR_s1_all"
  type: "Concat"
  bottom: "output_SR_s1_bp0"
  bottom: "output_SR_s1_bp6"
  top: "concat_SR_s1_all"
}

layer {
  name: "conv1_SR_s1_all"
  type: "Convolution"
  bottom: "concat_SR_s1_all"
  top: "conv1_SR_s1_all"
  param {
    lr_mult: 1
    name: "conv1_SR_s1_all_w"
  }
  param {
    lr_mult: 1
    name: "conv1_SR_s1_all_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu_conv1_SR_s1_all"
  type: "PReLU"
  bottom: "conv1_SR_s1_all"
  top: "conv1_SR_s1_all"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "data_s1"
  type: "Convolution"
  bottom: "data_upscale"
  top: "data_s1"
  param {
    lr_mult: 1
    name: "data_s1_w"
  }
  param {
    lr_mult: 1
    name: "data_s1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise_SR_final_s1"
    type: "Eltwise"
    bottom: "conv1_SR_s1_all"
    bottom: "data_s1"
    top: "SR_concate_s1"
    eltwise_param {
        operation: 1
    }
}

################################# s2##############################
######################### s2 bp1 #########################

layer {
  name: "conv1_SR_s2_bp1"
  type: "Convolution"
  bottom: "SR_concate_s1"
  top: "conv1_SR_s2_bp1"
  param {
    lr_mult: 1
    name: "conv1_bp1_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp1_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s2_bp1"
  type: "PReLU"
  bottom: "conv1_SR_s2_bp1"
  top: "conv1_SR_s2_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s2_bp1"
  type: "Deconvolution"
  bottom: "conv1_SR_s2_bp1"
  top: "conv2_SR_s2_bp1"
  param {
    lr_mult: 1
    name: "conv2_bp1_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp1_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s2_bp1"
  type: "PReLU"
  bottom: "conv2_SR_s2_bp1"
  top: "conv2_SR_s2_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s2_bp0"
  type: "Convolution"
  bottom: "SR_concate_s1"
  top: "local_output_SR_s2_bp0"
  param {
    lr_mult: 1
    name: "local_output_bp0_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_output_bp0_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s2_bp1"
    type: "Eltwise"
    bottom: "local_output_SR_s2_bp0"
    bottom: "conv2_SR_s2_bp1"
    top: "eltwise1_SR_s2_bp1"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s2_bp1"
  type: "Convolution"
  bottom: "eltwise1_SR_s2_bp1"
  top: "conv3_SR_s2_bp1"
  param {
    lr_mult: 1
    name: "conv3_bp1_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp1_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s2_bp1"
  type: "PReLU"
  bottom: "conv3_SR_s2_bp1"
  top: "conv3_SR_s2_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s2_bp1"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp1"
  top: "local_conv1_SR_s2_bp1"
  param {
    lr_mult: 1
    name: "local_conv1_bp1_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp1_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s2_bp1"
    type: "Eltwise"
    bottom: "local_conv1_SR_s2_bp1"
    bottom: "conv3_SR_s2_bp1"
    top: "output_SR_s2_bp1"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s2_bp1"
  type: "PReLU"
  bottom: "output_SR_s2_bp1"
  top: "output_SR_s2_bp1"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s1_bp1 & SR_s2_bp1 #########################

layer {
  name: "concat_SR_s1_bp1_SR_s2_bp1"
  type: "Concat"
  bottom: "output_SR_s1_bp1"
  bottom: "output_SR_s2_bp1"
  top: "concat_SR_s1_bp1_SR_s2_bp1"
}

######################### s2 bp2 #########################
layer {
  name: "conv1_SR_s2_bp2"
  type: "Convolution"
  bottom: "concat_SR_s1_bp1_SR_s2_bp1"
  top: "conv1_SR_s2_bp2"
  param {
    lr_mult: 1
    name: "conv1_bp2_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp2_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s2_bp2"
  type: "PReLU"
  bottom: "conv1_SR_s2_bp2"
  top: "conv1_SR_s2_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s2_bp2"
  type: "Deconvolution"
  bottom: "conv1_SR_s2_bp2"
  top: "conv2_SR_s2_bp2"
  param {
    lr_mult: 1
    name: "conv2_bp2_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp2_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s2_bp2"
  type: "PReLU"
  bottom: "conv2_SR_s2_bp2"
  top: "conv2_SR_s2_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s2_bp1"
  type: "Convolution"
  bottom: "concat_SR_s1_bp1_SR_s2_bp1"
  top: "local_output_SR_s2_bp1"
  param {
    lr_mult: 1
    name: "local_output_bp1_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_output_bp1_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s2_bp2"
    type: "Eltwise"
    bottom: "local_output_SR_s2_bp1"
    bottom: "conv2_SR_s2_bp2"
    top: "eltwise1_SR_s2_bp2"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s2_bp2"
  type: "Convolution"
  bottom: "eltwise1_SR_s2_bp2"
  top: "conv3_SR_s2_bp2"
  param {
    lr_mult: 1
    name: "conv3_bp2_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp2_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s2_bp2"
  type: "PReLU"
  bottom: "conv3_SR_s2_bp2"
  top: "conv3_SR_s2_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s2_bp2"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp2"
  top: "local_conv1_SR_s2_bp2"
  param {
    lr_mult: 1
    name: "local_conv1_bp2_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp2_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s2_bp2"
    type: "Eltwise"
    bottom: "local_conv1_SR_s2_bp2"
    bottom: "conv3_SR_s2_bp2"
    top: "output_SR_s2_bp2"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s2_bp2"
  type: "PReLU"
  bottom: "output_SR_s2_bp2"
  top: "output_SR_s2_bp2"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s1_bp2 & SR_s2_bp2 #########################

layer {
  name: "concat_SR_s1_bp2_SR_s2_bp2"
  type: "Concat"
  bottom: "output_SR_s1_bp2"
  bottom: "output_SR_s2_bp2"
  top: "concat_SR_s1_bp2_SR_s2_bp2"
}

######################### s2 bp3 #########################

layer {
  name: "conv1_SR_s2_bp3"
  type: "Convolution"
  bottom: "concat_SR_s1_bp2_SR_s2_bp2"
  top: "conv1_SR_s2_bp3"
  param {
    lr_mult: 1
    name: "conv1_s2_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s2_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s2_bp3"
  type: "PReLU"
  bottom: "conv1_SR_s2_bp3"
  top: "conv1_SR_s2_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s2_bp3"
  type: "Deconvolution"
  bottom: "conv1_SR_s2_bp3"
  top: "conv2_SR_s2_bp3"
  param {
    lr_mult: 1
    name: "conv2_s2_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s2_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s2_bp3"
  type: "PReLU"
  bottom: "conv2_SR_s2_bp3"
  top: "conv2_SR_s2_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s2_bp2"
  type: "Convolution"
  bottom: "concat_SR_s1_bp2_SR_s2_bp2"
  top: "local_output_SR_s2_bp2"
  param {
    lr_mult: 1
    name: "local_output_s2_bp2_w"
  }
  param {
    lr_mult: 1
    name: "local_output_s2_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s2_bp3"
    type: "Eltwise"
    bottom: "local_output_SR_s2_bp2"
    bottom: "conv2_SR_s2_bp3"
    top: "eltwise1_SR_s2_bp3"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s2_bp3"
  type: "Convolution"
  bottom: "eltwise1_SR_s2_bp3"
  top: "conv3_SR_s2_bp3"
  param {
    lr_mult: 1
    name: "conv3_s2_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s2_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s2_bp3"
  type: "PReLU"
  bottom: "conv3_SR_s2_bp3"
  top: "conv3_SR_s2_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s2_bp3"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp3"
  top: "local_conv1_SR_s2_bp3"
  param {
    lr_mult: 1
    name: "local_conv1_s2_bp3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s2_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s2_bp3"
    type: "Eltwise"
    bottom: "local_conv1_SR_s2_bp3"
    bottom: "conv3_SR_s2_bp3"
    top: "output_SR_s2_bp3"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s2_bp3"
  type: "PReLU"
  bottom: "output_SR_s2_bp3"
  top: "output_SR_s2_bp3"
  prelu_param {
    channel_shared: 1
  }
}

######################### s2 bp3 data duplicate #########################
layer {
  name: "data_duplicate_s2"
  type: "Convolution"
  bottom: "data"
  top: "data_duplicate_s2"
  param {
    lr_mult: 1
    name: "data_duplicate_s2_w"
  }
  param {
    lr_mult: 1
    name: "data_duplicate_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      value: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "residual_SR_s2_bp3"
    type: "Eltwise"
    bottom: "output_SR_s2_bp3"
    bottom: "data_duplicate_s2"
    top: "residual_SR_s2_bp3"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "concat_SR_s1_bp3_SR_s2_bp3"
  type: "Concat"
  bottom: "residual_SR_s1_bp3"
  bottom: "residual_SR_s2_bp3"
  top: "concat_SR_s1_bp3_SR_s2_bp3"
}

######################### SR_s2_bp4 #########################
layer {
  name: "conv1_SR_s2_bp4"
  type: "Deconvolution"
  bottom: "concat_SR_s1_bp3_SR_s2_bp3"
  top: "conv1_SR_s2_bp4"
  param {
    lr_mult: 1
    name: "conv1_bp4_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp4_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s2_bp4"
  type: "PReLU"
  bottom: "conv1_SR_s2_bp4"
  top: "conv1_SR_s2_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s2_bp4"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp4"
  top: "conv2_SR_s2_bp4"
  param {
    lr_mult: 1
    name: "conv2_bp4_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp4_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s2_bp4"
  type: "PReLU"
  bottom: "conv2_SR_s2_bp4"
  top: "conv2_SR_s2_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s2_bp3"
  type: "Convolution"
  bottom: "concat_SR_s1_bp3_SR_s2_bp3"
  top: "local_output_SR_s2_bp3"
  param {
    lr_mult: 1
    name: "local_output_bp3_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_output_bp3_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s2_bp4"
    type: "Eltwise"
    bottom: "local_output_SR_s2_bp3"
    bottom: "conv2_SR_s2_bp4"
    top: "eltwise1_SR_s2_bp4"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s2_bp4"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s2_bp4"
  top: "conv3_SR_s2_bp4"
  param {
    lr_mult: 1
    name: "conv3_bp4_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp4_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s2_bp4"
  type: "PReLU"
  bottom: "conv3_SR_s2_bp4"
  top: "conv3_SR_s2_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s2_bp4"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp4"
  top: "local_conv1_SR_s2_bp4"
  param {
    lr_mult: 1
    name: "local_conv1_bp4_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp4_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s2_bp4"
    type: "Eltwise"
    bottom: "local_conv1_SR_s2_bp4"
    bottom: "conv3_SR_s2_bp4"
    top: "output_SR_s2_bp4"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s2_bp4"
  type: "PReLU"
  bottom: "output_SR_s2_bp4"
  top: "output_SR_s2_bp4"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s2_bp2 & SR_s2_bp4 #########################
layer {
  name: "global_output_SR_s2_bp2"
  type: "Convolution"
  bottom: "output_SR_s2_bp2"
  top: "global_output_SR_s2_bp2"
  param {
    lr_mult: 1
    name: "global_output_bp2_s2_w"
  }
  param {
    lr_mult: 1
    name: "global_output_bp2_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "concat_SR_s2_bp2_SR_s2_bp4"
  type: "Concat"
  bottom: "global_output_SR_s1_bp2"
  bottom: "global_output_SR_s2_bp2"
  bottom: "output_SR_s2_bp4"
  top: "concat_SR_s2_bp2_SR_s2_bp4"
}

######################### SR_s2_bp5 #########################
layer {
  name: "conv1_SR_s2_bp5"
  type: "Deconvolution"
  bottom: "concat_SR_s2_bp2_SR_s2_bp4"
  top: "conv1_SR_s2_bp5"
  param {
    lr_mult: 1
    name: "conv1_bp5_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp5_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s2_bp5"
  type: "PReLU"
  bottom: "conv1_SR_s2_bp5"
  top: "conv1_SR_s2_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s2_bp5"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp5"
  top: "conv2_SR_s2_bp5"
  param {
    lr_mult: 1
    name: "conv2_bp5_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp5_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s2_bp5"
  type: "PReLU"
  bottom: "conv2_SR_s2_bp5"
  top: "conv2_SR_s2_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_concat_SR_s2_bp2_SR_s2_bp4"
  type: "Convolution"
  bottom: "concat_SR_s2_bp2_SR_s2_bp4"
  top: "local_concat_SR_s2_bp2_SR_s2_bp4"
  param {
    lr_mult: 1
    name: "local_concat_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_concat_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s2_bp5"
    type: "Eltwise"
    bottom: "local_concat_SR_s2_bp2_SR_s2_bp4"
    bottom: "conv2_SR_s2_bp5"
    top: "eltwise1_SR_s2_bp5"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s2_bp5"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s2_bp5"
  top: "conv3_SR_s2_bp5"
  param {
    lr_mult: 1
    name: "conv3_bp5_s2_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp5_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s2_bp5"
  type: "PReLU"
  bottom: "conv3_SR_s2_bp5"
  top: "conv3_SR_s2_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s2_bp5"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp5"
  top: "local_conv1_SR_s2_bp5"
  param {
    lr_mult: 1
    name: "local_conv1_bp5_s2_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp5_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s2_bp5"
    type: "Eltwise"
    bottom: "local_conv1_SR_s2_bp5"
    bottom: "conv3_SR_s2_bp5"
    top: "output_SR_s2_bp5"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s2_bp5"
  type: "PReLU"
  bottom: "output_SR_s2_bp5"
  top: "output_SR_s2_bp5"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s2_bp1 & SR_s2_bp5 #########################
layer {
  name: "global_output_SR_s2_bp1"
  type: "Convolution"
  bottom: "output_SR_s2_bp1"
  top: "global_output_SR_s2_bp1"
  param {
    lr_mult: 1
    name: "global_output_s2_bp1_w"
  }
  param {
    lr_mult: 1
    name: "global_output_s2_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "concat_SR_s2_bp1_SR_s2_bp5"
  type: "Concat"
  bottom: "global_output_SR_s1_bp1"
  bottom: "global_output_SR_s2_bp1"
  bottom: "output_SR_s2_bp5"
  top: "concat_SR_s2_bp1_SR_s2_bp5"
}

######################### SR_s2_bp6 #########################
layer {
  name: "conv1_SR_s2_bp6"
  type: "Deconvolution"
  bottom: "concat_SR_s2_bp1_SR_s2_bp5"
  top: "conv1_SR_s2_bp6"
  param {
    lr_mult: 1
    name: "conv1_s2_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s2_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s2_bp6"
  type: "PReLU"
  bottom: "conv1_SR_s2_bp6"
  top: "conv1_SR_s2_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s2_bp6"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp6"
  top: "conv2_SR_s2_bp6"
  param {
    lr_mult: 1
    name: "conv2_s2_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s2_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s2_bp6"
  type: "PReLU"
  bottom: "conv2_SR_s2_bp6"
  top: "conv2_SR_s2_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s2_bp5"
  type: "Convolution"
  bottom: "concat_SR_s2_bp1_SR_s2_bp5"
  top: "local_output_SR_s2_bp5"
  param {
    lr_mult: 1
    name: "local_output_SR_s2_bp5_w"
  }
  param {
    lr_mult: 1
    name: "local_output_SR_s2_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s2_bp6"
    type: "Eltwise"
    bottom: "local_output_SR_s2_bp5"
    bottom: "conv2_SR_s2_bp6"
    top: "eltwise1_SR_s2_bp6"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s2_bp6"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s2_bp6"
  top: "conv3_SR_s2_bp6"
  param {
    lr_mult: 1
    name: "conv3_s2_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s2_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s2_bp6"
  type: "PReLU"
  bottom: "conv3_SR_s2_bp6"
  top: "conv3_SR_s2_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s2_bp6"
  type: "Convolution"
  bottom: "conv1_SR_s2_bp6"
  top: "local_conv1_SR_s2_bp6"
  param {
    lr_mult: 1
    name: "local_conv1_s2_bp6_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s2_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s2_bp6"
    type: "Eltwise"
    bottom: "local_conv1_SR_s2_bp6"
    bottom: "conv3_SR_s2_bp6"
    top: "output_SR_s2_bp6"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s2_bp6"
  type: "PReLU"
  bottom: "output_SR_s2_bp6"
  top: "output_SR_s2_bp6"
  prelu_param {
    channel_shared: 1
  }
}

######################### concate SR_s2 all #########################
layer {
  name: "concat_SR_s2_all"
  type: "Concat"
  bottom: "conv1_SR_s1_all"
  bottom: "output_SR_s2_bp6"
  top: "concat_SR_s2_all"
}

layer {
  name: "conv1_SR_s2_all"
  type: "Convolution"
  bottom: "concat_SR_s2_all"
  top: "conv1_SR_s2_all"
  param {
    lr_mult: 1
    name: "conv1_SR_s2_all_w"
  }
  param {
    lr_mult: 1
    name: "conv1_SR_s2_all_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu_conv1_SR_s2_all"
  type: "PReLU"
  bottom: "conv1_SR_s2_all"
  top: "conv1_SR_s2_all"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "data_s2"
  type: "Convolution"
  bottom: "SR_concate_s1"
  top: "data_s2"
  param {
    lr_mult: 1
    name: "data_s2_w"
  }
  param {
    lr_mult: 1
    name: "data_s2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise_SR_final_s2"
    type: "Eltwise"
    bottom: "conv1_SR_s2_all"
    bottom: "data_s2"
    top: "SR_concate_s2"
    eltwise_param {
        operation: 1
    }
}


############################ s3 ##########################
######################### s3 bp1 #########################

layer {
  name: "conv1_SR_s3_bp1"
  type: "Convolution"
  bottom: "SR_concate_s2"
  top: "conv1_SR_s3_bp1"
  param {
    lr_mult: 1
    name: "conv1_bp1_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp1_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s3_bp1"
  type: "PReLU"
  bottom: "conv1_SR_s3_bp1"
  top: "conv1_SR_s3_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s3_bp1"
  type: "Deconvolution"
  bottom: "conv1_SR_s3_bp1"
  top: "conv2_SR_s3_bp1"
  param {
    lr_mult: 1
    name: "conv2_bp1_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp1_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s3_bp1"
  type: "PReLU"
  bottom: "conv2_SR_s3_bp1"
  top: "conv2_SR_s3_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s3_bp0"
  type: "Convolution"
  bottom: "SR_concate_s2"
  top: "local_output_SR_s3_bp0"
  param {
    lr_mult: 1
    name: "local_output_bp0_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_output_bp0_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s3_bp1"
    type: "Eltwise"
    bottom: "local_output_SR_s3_bp0"
    bottom: "conv2_SR_s3_bp1"
    top: "eltwise1_SR_s3_bp1"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s3_bp1"
  type: "Convolution"
  bottom: "eltwise1_SR_s3_bp1"
  top: "conv3_SR_s3_bp1"
  param {
    lr_mult: 1
    name: "conv3_bp1_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp1_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s3_bp1"
  type: "PReLU"
  bottom: "conv3_SR_s3_bp1"
  top: "conv3_SR_s3_bp1"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s3_bp1"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp1"
  top: "local_conv1_SR_s3_bp1"
  param {
    lr_mult: 1
    name: "local_conv1_bp1_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp1_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s3_bp1"
    type: "Eltwise"
    bottom: "local_conv1_SR_s3_bp1"
    bottom: "conv3_SR_s3_bp1"
    top: "output_SR_s3_bp1"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s3_bp1"
  type: "PReLU"
  bottom: "output_SR_s3_bp1"
  top: "output_SR_s3_bp1"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s2_bp1 & SR_s3_bp1 #########################

layer {
  name: "concat_SR_s2_bp1_SR_s3_bp1"
  type: "Concat"
  bottom: "output_SR_s1_bp1"
  bottom: "output_SR_s2_bp1"
  bottom: "output_SR_s3_bp1"
  top: "concat_SR_s2_bp1_SR_s3_bp1"
}

######################### s3 bp2 #########################
layer {
  name: "conv1_SR_s3_bp2"
  type: "Convolution"
  bottom: "concat_SR_s2_bp1_SR_s3_bp1"
  top: "conv1_SR_s3_bp2"
  param {
    lr_mult: 1
    name: "conv1_bp2_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp2_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s3_bp2"
  type: "PReLU"
  bottom: "conv1_SR_s3_bp2"
  top: "conv1_SR_s3_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s3_bp2"
  type: "Deconvolution"
  bottom: "conv1_SR_s3_bp2"
  top: "conv2_SR_s3_bp2"
  param {
    lr_mult: 1
    name: "conv2_bp2_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp2_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s3_bp2"
  type: "PReLU"
  bottom: "conv2_SR_s3_bp2"
  top: "conv2_SR_s3_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s3_bp1"
  type: "Convolution"
  bottom: "concat_SR_s2_bp1_SR_s3_bp1"
  top: "local_output_SR_s3_bp1"
  param {
    lr_mult: 1
    name: "local_output_bp1_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_output_bp1_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s3_bp2"
    type: "Eltwise"
    bottom: "local_output_SR_s3_bp1"
    bottom: "conv2_SR_s3_bp2"
    top: "eltwise1_SR_s3_bp2"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s3_bp2"
  type: "Convolution"
  bottom: "eltwise1_SR_s3_bp2"
  top: "conv3_SR_s3_bp2"
  param {
    lr_mult: 1
    name: "conv3_bp2_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp2_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s3_bp2"
  type: "PReLU"
  bottom: "conv3_SR_s3_bp2"
  top: "conv3_SR_s3_bp2"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s3_bp2"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp2"
  top: "local_conv1_SR_s3_bp2"
  param {
    lr_mult: 1
    name: "local_conv1_bp2_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp2_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s3_bp2"
    type: "Eltwise"
    bottom: "local_conv1_SR_s3_bp2"
    bottom: "conv3_SR_s3_bp2"
    top: "output_SR_s3_bp2"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s3_bp2"
  type: "PReLU"
  bottom: "output_SR_s3_bp2"
  top: "output_SR_s3_bp2"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s2_bp2 & SR_s3_bp2 #########################

layer {
  name: "concat_SR_s2_bp2_SR_s3_bp2"
  type: "Concat"
  bottom: "output_SR_s1_bp2"
  bottom: "output_SR_s2_bp2"
  bottom: "output_SR_s3_bp2"
  top: "concat_SR_s2_bp2_SR_s3_bp2"
}

######################### s3 bp3 #########################

layer {
  name: "conv1_SR_s3_bp3"
  type: "Convolution"
  bottom: "concat_SR_s2_bp2_SR_s3_bp2"
  top: "conv1_SR_s3_bp3"
  param {
    lr_mult: 1
    name: "conv1_s3_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s3_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s3_bp3"
  type: "PReLU"
  bottom: "conv1_SR_s3_bp3"
  top: "conv1_SR_s3_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s3_bp3"
  type: "Deconvolution"
  bottom: "conv1_SR_s3_bp3"
  top: "conv2_SR_s3_bp3"
  param {
    lr_mult: 1
    name: "conv2_s3_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s3_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s3_bp3"
  type: "PReLU"
  bottom: "conv2_SR_s3_bp3"
  top: "conv2_SR_s3_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s3_bp2"
  type: "Convolution"
  bottom: "concat_SR_s2_bp2_SR_s3_bp2"
  top: "local_output_SR_s3_bp2"
  param {
    lr_mult: 1
    name: "local_output_s3_bp2_w"
  }
  param {
    lr_mult: 1
    name: "local_output_s3_bp2_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s3_bp3"
    type: "Eltwise"
    bottom: "local_output_SR_s3_bp2"
    bottom: "conv2_SR_s3_bp3"
    top: "eltwise1_SR_s3_bp3"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s3_bp3"
  type: "Convolution"
  bottom: "eltwise1_SR_s3_bp3"
  top: "conv3_SR_s3_bp3"
  param {
    lr_mult: 1
    name: "conv3_s3_bp3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s3_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s3_bp3"
  type: "PReLU"
  bottom: "conv3_SR_s3_bp3"
  top: "conv3_SR_s3_bp3"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s3_bp3"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp3"
  top: "local_conv1_SR_s3_bp3"
  param {
    lr_mult: 1
    name: "local_conv1_s3_bp3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s3_bp3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s3_bp3"
    type: "Eltwise"
    bottom: "local_conv1_SR_s3_bp3"
    bottom: "conv3_SR_s3_bp3"
    top: "output_SR_s3_bp3"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s3_bp3"
  type: "PReLU"
  bottom: "output_SR_s3_bp3"
  top: "output_SR_s3_bp3"
  prelu_param {
    channel_shared: 1
  }
}

######################### s3 bp3 data duplicate #########################
layer {
  name: "data_duplicate_s3"
  type: "Convolution"
  bottom: "data"
  top: "data_duplicate_s3"
  param {
    lr_mult: 1
    name: "data_duplicate_s3_w"
  }
  param {
    lr_mult: 1
    name: "data_duplicate_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      value: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "residual_SR_s3_bp3"
    type: "Eltwise"
    bottom: "output_SR_s3_bp3"
    bottom: "data_duplicate_s3"
    top: "residual_SR_s3_bp3"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "concat_SR_s2_bp3_SR_s3_bp3"
  type: "Concat"
  bottom: "concat_SR_s1_bp3_SR_s2_bp3"
  bottom: "residual_SR_s3_bp3"
  top: "concat_SR_s2_bp3_SR_s3_bp3"
}

######################### SR_s3_bp4 #########################
layer {
  name: "conv1_SR_s3_bp4"
  type: "Deconvolution"
  bottom: "concat_SR_s2_bp3_SR_s3_bp3"
  top: "conv1_SR_s3_bp4"
  param {
    lr_mult: 1
    name: "conv1_bp4_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp4_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s3_bp4"
  type: "PReLU"
  bottom: "conv1_SR_s3_bp4"
  top: "conv1_SR_s3_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s3_bp4"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp4"
  top: "conv2_SR_s3_bp4"
  param {
    lr_mult: 1
    name: "conv2_bp4_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp4_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s3_bp4"
  type: "PReLU"
  bottom: "conv2_SR_s3_bp4"
  top: "conv2_SR_s3_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s3_bp3"
  type: "Convolution"
  bottom: "concat_SR_s2_bp3_SR_s3_bp3"
  top: "local_output_SR_s3_bp3"
  param {
    lr_mult: 1
    name: "local_output_bp3_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_output_bp3_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s3_bp4"
    type: "Eltwise"
    bottom: "local_output_SR_s3_bp3"
    bottom: "conv2_SR_s3_bp4"
    top: "eltwise1_SR_s3_bp4"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s3_bp4"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s3_bp4"
  top: "conv3_SR_s3_bp4"
  param {
    lr_mult: 1
    name: "conv3_bp4_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp4_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s3_bp4"
  type: "PReLU"
  bottom: "conv3_SR_s3_bp4"
  top: "conv3_SR_s3_bp4"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s3_bp4"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp4"
  top: "local_conv1_SR_s3_bp4"
  param {
    lr_mult: 1
    name: "local_conv1_bp4_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp4_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s3_bp4"
    type: "Eltwise"
    bottom: "local_conv1_SR_s3_bp4"
    bottom: "conv3_SR_s3_bp4"
    top: "output_SR_s3_bp4"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s3_bp4"
  type: "PReLU"
  bottom: "output_SR_s3_bp4"
  top: "output_SR_s3_bp4"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s3_bp2 & SR_s3_bp4 #########################
layer {
  name: "global_output_SR_s3_bp2"
  type: "Convolution"
  bottom: "output_SR_s3_bp2"
  top: "global_output_SR_s3_bp2"
  param {
    lr_mult: 1
    name: "global_output_bp2_s3_w"
  }
  param {
    lr_mult: 1
    name: "global_output_bp2_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "concat_SR_s3_bp2_SR_s3_bp4"
  type: "Concat"
  bottom: "global_output_SR_s1_bp2"
  bottom: "global_output_SR_s2_bp2"
  bottom: "global_output_SR_s3_bp2"
  bottom: "output_SR_s3_bp4"
  top: "concat_SR_s3_bp2_SR_s3_bp4"
}

######################### SR_s3_bp5 #########################
layer {
  name: "conv1_SR_s3_bp5"
  type: "Deconvolution"
  bottom: "concat_SR_s3_bp2_SR_s3_bp4"
  top: "conv1_SR_s3_bp5"
  param {
    lr_mult: 1
    name: "conv1_bp5_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv1_bp5_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s3_bp5"
  type: "PReLU"
  bottom: "conv1_SR_s3_bp5"
  top: "conv1_SR_s3_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s3_bp5"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp5"
  top: "conv2_SR_s3_bp5"
  param {
    lr_mult: 1
    name: "conv2_bp5_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv2_bp5_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s3_bp5"
  type: "PReLU"
  bottom: "conv2_SR_s3_bp5"
  top: "conv2_SR_s3_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_concat_SR_s3_bp2_SR_s3_bp4"
  type: "Convolution"
  bottom: "concat_SR_s3_bp2_SR_s3_bp4"
  top: "local_concat_SR_s3_bp2_SR_s3_bp4"
  param {
    lr_mult: 1
    name: "local_concat_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_concat_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s3_bp5"
    type: "Eltwise"
    bottom: "local_concat_SR_s3_bp2_SR_s3_bp4"
    bottom: "conv2_SR_s3_bp5"
    top: "eltwise1_SR_s3_bp5"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s3_bp5"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s3_bp5"
  top: "conv3_SR_s3_bp5"
  param {
    lr_mult: 1
    name: "conv3_bp5_s3_w"
  }
  param {
    lr_mult: 1
    name: "conv3_bp5_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s3_bp5"
  type: "PReLU"
  bottom: "conv3_SR_s3_bp5"
  top: "conv3_SR_s3_bp5"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s3_bp5"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp5"
  top: "local_conv1_SR_s3_bp5"
  param {
    lr_mult: 1
    name: "local_conv1_bp5_s3_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_bp5_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s3_bp5"
    type: "Eltwise"
    bottom: "local_conv1_SR_s3_bp5"
    bottom: "conv3_SR_s3_bp5"
    top: "output_SR_s3_bp5"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s3_bp5"
  type: "PReLU"
  bottom: "output_SR_s3_bp5"
  top: "output_SR_s3_bp5"
  prelu_param {
    channel_shared: 1
  }
}

######################### SR_s3_bp1 & SR_s3_bp5 #########################
layer {
  name: "global_output_SR_s3_bp1"
  type: "Convolution"
  bottom: "output_SR_s3_bp1"
  top: "global_output_SR_s3_bp1"
  param {
    lr_mult: 1
    name: "global_output_s3_bp1_w"
  }
  param {
    lr_mult: 1
    name: "global_output_s3_bp1_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "concat_SR_s3_bp1_SR_s3_bp5"
  type: "Concat"
  bottom: "global_output_SR_s1_bp1"
  bottom: "global_output_SR_s2_bp1"
  bottom: "global_output_SR_s3_bp1"
  bottom: "output_SR_s3_bp5"
  top: "concat_SR_s3_bp1_SR_s3_bp5"
}

######################### SR_s3_bp6 #########################
layer {
  name: "conv1_SR_s3_bp6"
  type: "Deconvolution"
  bottom: "concat_SR_s3_bp1_SR_s3_bp5"
  top: "conv1_SR_s3_bp6"
  param {
    lr_mult: 1
    name: "conv1_s3_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv1_s3_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu1_SR_s3_bp6"
  type: "PReLU"
  bottom: "conv1_SR_s3_bp6"
  top: "conv1_SR_s3_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "conv2_SR_s3_bp6"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp6"
  top: "conv2_SR_s3_bp6"
  param {
    lr_mult: 1
    name: "conv2_s3_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv2_s3_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu2_SR_s3_bp6"
  type: "PReLU"
  bottom: "conv2_SR_s3_bp6"
  top: "conv2_SR_s3_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_output_SR_s3_bp5"
  type: "Convolution"
  bottom: "concat_SR_s3_bp1_SR_s3_bp5"
  top: "local_output_SR_s3_bp5"
  param {
    lr_mult: 1
    name: "local_output_SR_s3_bp5_w"
  }
  param {
    lr_mult: 1
    name: "local_output_SR_s3_bp5_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise1_SR_s3_bp6"
    type: "Eltwise"
    bottom: "local_output_SR_s3_bp5"
    bottom: "conv2_SR_s3_bp6"
    top: "eltwise1_SR_s3_bp6"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "conv3_SR_s3_bp6"
  type: "Deconvolution"
  bottom: "eltwise1_SR_s3_bp6"
  top: "conv3_SR_s3_bp6"
  param {
    lr_mult: 1
    name: "conv3_s3_bp6_w"
  }
  param {
    lr_mult: 1
    name: "conv3_s3_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 6
    stride: 2
    pad: 2
    weight_filler {
      type: "gaussian"
      std: 0.0417
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu3_SR_s3_bp6"
  type: "PReLU"
  bottom: "conv3_SR_s3_bp6"
  top: "conv3_SR_s3_bp6"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "local_conv1_SR_s3_bp6"
  type: "Convolution"
  bottom: "conv1_SR_s3_bp6"
  top: "local_conv1_SR_s3_bp6"
  param {
    lr_mult: 1
    name: "local_conv1_s3_bp6_w"
  }
  param {
    lr_mult: 1
    name: "local_conv1_s3_bp6_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise2_SR_s3_bp6"
    type: "Eltwise"
    bottom: "local_conv1_SR_s3_bp6"
    bottom: "conv3_SR_s3_bp6"
    top: "output_SR_s3_bp6"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "relu_output_SR_s3_bp6"
  type: "PReLU"
  bottom: "output_SR_s3_bp6"
  top: "output_SR_s3_bp6"
  prelu_param {
    channel_shared: 1
  }
}

######################### concate SR_s3 all #########################
layer {
  name: "concat_SR_s3_all"
  type: "Concat"
  bottom: "conv1_SR_s1_all"
  bottom: "conv1_SR_s2_all"
  bottom: "output_SR_s3_bp6"
  top: "concat_SR_s3_all"
}

layer {
  name: "conv1_SR_s3_all"
  type: "Convolution"
  bottom: "concat_SR_s3_all"
  top: "conv1_SR_s3_all"
  param {
    lr_mult: 1
    name: "conv1_SR_s3_all_w"
  }
  param {
    lr_mult: 1
    name: "conv1_SR_s3_all_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.0833
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
  name: "relu_conv1_SR_s3_all"
  type: "PReLU"
  bottom: "conv1_SR_s3_all"
  top: "conv1_SR_s3_all"
  prelu_param {
    channel_shared: 1
  }
}

layer {
  name: "data_s3"
  type: "Convolution"
  bottom: "SR_concate_s2"
  top: "data_s3"
  param {
    lr_mult: 1
    name: "data_s3_w"
  }
  param {
    lr_mult: 1
    name: "data_s3_b"
  }
  convolution_param {
    num_output: 32
    kernel_size: 1
    stride: 1
    pad: 0
    weight_filler {
      type: "gaussian"
      std: 0.25
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

layer {
    name: "eltwise_SR_final_s3"
    type: "Eltwise"
    bottom: "conv1_SR_s3_all"
    bottom: "data_s3"
    top: "SR_concate_s3"
    eltwise_param {
        operation: 1
    }
}

layer {
  name: "SR_final_s3"
  type: "Convolution"
  bottom: "SR_concate_s3"
  top: "SR_final_s3"
  param {
    lr_mult: 1
    name: "SR_final_s3_w"
  }
  param {
    lr_mult: 1
    name: "SR_final_s3_b"
  }
  convolution_param {
    num_output: 3
    kernel_size: 3
    stride: 1
    pad: 1
    weight_filler {
      type: "gaussian"
      std: 0.2722
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}

######################### SR MAE LOSS #########################
layer {
    name: "eltwise_loss"
    type: "Eltwise"
    bottom: "label"
    bottom: "SR_final_s3"
    top: "diff"
    eltwise_param {
        operation: SUM
        coeff: 1
        coeff: -1
    }
}

layer {
  name: "layer"
  bottom: "diff"
  top: "abs"
  type: "AbsVal"
}

layer {
  name: "MAE_loss"
  type: "InnerProduct"
  param { lr_mult: 0 decay_mult: 0 }
  param { lr_mult: 0 decay_mult: 0 }
  inner_product_param {
    num_output: 1
    weight_filler {
      type: "constant"
      value: 1    
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
  bottom: "abs"
  top: "MAE_loss"
  loss_weight: 1
}