My_deploy.prototxt

#######################################
#        LiteFlowNet CVPR 2018        #
#                by                   #
#           T.-W. Hui, CUHK           #
#         Edited by Amin Ullah        #
#######################################
input: "img0"
input: "img1"

  input_shape {
  dim: 1
  dim: 3
  dim: 256
  dim: 320
}

  input_shape {
  dim: 1
  dim: 3
  dim: 256
  dim: 320
}

#######################################
#           Pre-processing            #
#######################################
layer {
  name: "Eltwise1"
  type: "Eltwise"
  bottom: "img0"
  top: "img0s"
  eltwise_param {
    operation: SUM
    coeff: 0.00392156862745
  }
}
layer {
  name: "Eltwise2"
  type: "Eltwise"
  bottom: "img1"
  top: "img1s"
  eltwise_param {
    operation: SUM
    coeff: 0.00392156862745
  }
}
layer {
  name: "img0s_aug"
  type: "DataAugmentation"
  bottom: "img0s"
  top: "img0_nomean"
  augmentation_param {
    augment_during_test: true
    recompute_mean: 3000
    mean_per_pixel: false
  }
}
layer {
  name: "img1s_aug"
  type: "DataAugmentation"
  bottom: "img1s"
  top: "img1_nomean"
  augmentation_param {
    augment_during_test: true
    recompute_mean: 3000
    mean_per_pixel: false
  }
}
layer {
  name: "Resample1"
  type: "Resample"
  bottom: "img0_nomean"
  top: "img0_nomean_resize"
  resample_param {
    width: 320
    height: 256
    type: LINEAR
    antialias: true
  }
}
layer {
  name: "Resample2"
  type: "Resample"
  bottom: "img1_nomean"
  top: "img1_nomean_resize"
  resample_param {
    width: 320
    height: 256
    type: LINEAR
    antialias: true
  }
}
#######################################
#                NetC                 #
#######################################
layer {
  name: "conv1"
  type: "Convolution"
  bottom: "img0_nomean_resize"
  bottom: "img1_nomean_resize"
  top: "F0_L1"
  top: "F1_L1"
  convolution_param {
    num_output: 32
    pad: 3
    kernel_size: 7
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1a"
  type: "ReLU"
  bottom: "F0_L1"
  top: "F0_L1"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU1b"
  type: "ReLU"
  bottom: "F1_L1"
  top: "F1_L1"
  relu_param { negative_slope: 0.1 }
}

layer {
  name: "conv2_1"
  type: "Convolution"
  bottom: "F0_L1"
  bottom: "F1_L1"
  top: "F0_1_L2"
  top: "F1_1_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 2
    engine: CUDNN
  }
}
layer {
  name: "ReLU2a_1"
  type: "ReLU"
  bottom: "F0_1_L2"
  top: "F0_1_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU2b_1"
  type: "ReLU"
  bottom: "F1_1_L2"
  top: "F1_1_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_2"
  type: "Convolution"
  bottom: "F0_1_L2"
  bottom: "F1_1_L2"
  top: "F0_2_L2"
  top: "F1_2_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2a_2"
  type: "ReLU"
  bottom: "F0_2_L2"
  top: "F0_2_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU2b_2"
  type: "ReLU"
  bottom: "F1_2_L2"
  top: "F1_2_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_3"
  type: "Convolution"
  bottom: "F0_2_L2"
  bottom: "F1_2_L2"
  top: "F0_L2"
  top: "F1_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2a_3"
  type: "ReLU"
  bottom: "F0_L2"
  top: "F0_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU2b_3"
  type: "ReLU"
  bottom: "F1_L2"
  top: "F1_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_1"
  type: "Convolution"
  bottom: "F0_L2"
  bottom: "F1_L2"
  top: "F0_1_L3"
  top: "F1_1_L3"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 2
    engine: CUDNN
  }
}
layer {
  name: "ReLU3a_1"
  type: "ReLU"
  bottom: "F0_1_L3"
  top: "F0_1_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU3b_1"
  type: "ReLU"
  bottom: "F1_1_L3"
  top: "F1_1_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_2"
  type: "Convolution"
  bottom: "F0_1_L3"
  bottom: "F1_1_L3"
  top: "F0_L3"
  top: "F1_L3"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3a_2"
  type: "ReLU"
  bottom: "F0_L3"
  top: "F0_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU3b_2"
  type: "ReLU"
  bottom: "F1_L3"
  top: "F1_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_1"
  type: "Convolution"
  bottom: "F0_L3"
  bottom: "F1_L3"
  top: "F0_1_L4"
  top: "F1_1_L4"
  convolution_param {
    num_output: 96
    pad: 1
    kernel_size: 3
    stride: 2
    engine: CUDNN
  }
}
layer {
  name: "ReLU4a_1"
  type: "ReLU"
  bottom: "F0_1_L4"
  top: "F0_1_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU4b_1"
  type: "ReLU"
  bottom: "F1_1_L4"
  top: "F1_1_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_2"
  type: "Convolution"
  bottom: "F0_1_L4"
  bottom: "F1_1_L4"
  top: "F0_L4"
  top: "F1_L4"
  convolution_param {
    num_output: 96
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU4a_2"
  type: "ReLU"
  bottom: "F0_L4"
  top: "F0_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU4b_2"
  type: "ReLU"
  bottom: "F1_L4"
  top: "F1_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv5"
  type: "Convolution"
  bottom: "F0_L4"
  bottom: "F1_L4"
  top: "F0_L5"
  top: "F1_L5"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 2
    engine: CUDNN
  }
}
layer {
  name: "ReLU5a"
  type: "ReLU"
  bottom: "F0_L5"
  top: "F0_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU5b"
  type: "ReLU"
  bottom: "F1_L5"
  top: "F1_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv6"
  type: "Convolution"
  bottom: "F0_L5"
  bottom: "F1_L5"
  top: "F0_L6"
  top: "F1_L6"
  convolution_param {
    num_output: 192
    pad: 1
    kernel_size: 3
    stride: 2
    engine: CUDNN
  }
}
layer {
  name: "ReLU6a"
  type: "ReLU"
  bottom: "F0_L6"
  top: "F0_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "ReLU6b"
  type: "ReLU"
  bottom: "F1_L6"
  top: "F1_L6"
  relu_param { negative_slope: 0.1 }
}

#######################################
#         Downsample images           #
#######################################
layer {
  name: "Downsample_img0_aug_L2"
  type: "Downsample"
  bottom: "img0_nomean_resize"
  bottom: "F0_L2"
  top: "img0_aug_L2"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img1_aug_L2"
  type: "Downsample"
  bottom: "img1_nomean_resize"
  bottom: "F1_L2"
  top: "img1_aug_L2"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img0_aug_L3"
  type: "Downsample"
  bottom: "img0_nomean_resize"
  bottom: "F0_L3"
  top: "img0_aug_L3"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img1_aug_L3"
  type: "Downsample"
  bottom: "img1_nomean_resize"
  bottom: "F1_L3"
  top: "img1_aug_L3"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img0_aug_L4"
  type: "Downsample"
  bottom: "img0_nomean_resize"
  bottom: "F0_L4"
  top: "img0_aug_L4"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img1_aug_L4"
  type: "Downsample"
  bottom: "img1_nomean_resize"
  bottom: "F1_L4"
  top: "img1_aug_L4"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img0_aug_L5"
  type: "Downsample"
  bottom: "img0_aug_L4"
  bottom: "F0_L5"
  top: "img0_aug_L5"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img1_aug_L5"
  type: "Downsample"
  bottom: "img1_aug_L4"
  bottom: "F1_L5"
  top: "img1_aug_L5"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img0_aug_L6"
  type: "Downsample"
  bottom: "img0_aug_L4"
  bottom: "F0_L6"
  top: "img0_aug_L6"
  propagate_down: false
  propagate_down: false
}
layer {
  name: "Downsample_img1_aug_L6"
  type: "Downsample"
  bottom: "img1_aug_L4"
  bottom: "F1_L6"
  top: "img1_aug_L6"
  propagate_down: false
  propagate_down: false
}
#######################################
#             NetE-M: L6              #
#######################################
layer {
  name: "corr_L6"
  type: "Correlation"
  bottom: "F0_L6"
  bottom: "F1_L6"
  top: "corr_L6"
  correlation_param {
    pad: 3
    kernel_size: 1
    max_displacement: 3
    stride_1: 1
    stride_2: 1
  }
}
layer {
  name: "ReLU_corr_L6"
  type: "ReLU"
  bottom: "corr_L6"
  top: "corr_L6"
  relu_param {
    negative_slope: 0.1
  }
}
layer {
  name: "conv1_D1_L6"
  type: "Convolution"
  bottom: "corr_L6"
  top: "conv1_D1_L6"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D1_L6"
  type: "ReLU"
  bottom: "conv1_D1_L6"
  top: "conv1_D1_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D1_L6"
  type: "Convolution"
  bottom: "conv1_D1_L6"
  top: "conv2_D1_L6"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D1_L6"
  type: "ReLU"
  bottom: "conv2_D1_L6"
  top: "conv2_D1_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D1_L6"
  type: "Convolution"
  bottom: "conv2_D1_L6"
  top: "conv3_D1_L6"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D1_L6"
  type: "ReLU"
  bottom: "conv3_D1_L6"
  top: "conv3_D1_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D1_L6"
  type: "Convolution"
  bottom: "conv3_D1_L6"
  top: "scaled_flow_D1_L6"
  convolution_param {
    num_output: 2
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
#######################################
#             NetE-S: L6              #
#######################################
layer {
  name: "FlowUnscaling_L6_D2"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L6"
  top: "flow_D1_L6"
  eltwise_param {
    operation: SUM
    coeff: 0.625
  }
}
layer {
   name: "gxy_L6"
   type: "Grid"
   top: "gxy_L6"
   bottom: "flow_D1_L6"
   propagate_down: false
}
layer {
  name: "coords_D1_L6"
  type: "Eltwise"
  bottom: "flow_D1_L6"
  bottom: "gxy_L6"
  top: "coords_D1_L6"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_F1_L6"
  type: "Warp"
  bottom: "F1_L6"
  bottom: "coords_D1_L6"
  top: "warped_D1_F1_L6"
}
layer {
  name: "F_D2_L6"
  bottom: "F0_L6"
  bottom: "warped_D1_F1_L6"
  bottom: "scaled_flow_D1_L6"
  top: "F_D2_L6"
  type: "Concat"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_D2_L6"
  type: "Convolution"
  bottom: "F_D2_L6"
  top: "conv1_D2_L6"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D2_L6"
  type: "ReLU"
  bottom: "conv1_D2_L6"
  top: "conv1_D2_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D2_L6"
  type: "Convolution"
  bottom: "conv1_D2_L6"
  top: "conv2_D2_L6"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D2_L6"
  type: "ReLU"
  bottom: "conv2_D2_L6"
  top: "conv2_D2_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D2_L6"
  type: "Convolution"
  bottom: "conv2_D2_L6"
  top: "conv3_D2_L6"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D2_L6"
  type: "ReLU"
  bottom: "conv3_D2_L6"
  top: "conv3_D2_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D2_res_L6"
  type: "Convolution"
  bottom: "conv3_D2_L6"
  top: "scaled_flow_D2_res_L6"
  convolution_param {
    num_output: 2
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D2_L6"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L6"
  bottom: "scaled_flow_D2_res_L6"
  top: "scaled_flow_D2_L6"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-R: L6              #
#######################################
layer {
  name: "slice_scaled_flow_D_L6"
  type: "Slice"
  bottom: "scaled_flow_D2_L6"
  top: "scaled_flow_D_L6_x"
  top: "scaled_flow_D_L6_y"
  slice_param { axis: 1 slice_point: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L6_x"
  type: "Im2col"
  bottom: "scaled_flow_D_L6_x"
  top: "reshaped_scaled_flow_D_L6_x"
  convolution_param { pad: 1 kernel_size: 3 stride: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L6_y"
  type: "Im2col"
  bottom: "scaled_flow_D_L6_y"
  top: "reshaped_scaled_flow_D_L6_y"
  convolution_param { pad: 1 kernel_size: 3 stride: 1 }
}
layer {
  name: "mean_scaled_flow_D_L6_x"
  type: "Reduction"
  bottom: "scaled_flow_D_L6_x"
  top: "mean_scaled_flow_D_L6_x"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L6_x"
  type: "Bias"
  bottom: "scaled_flow_D_L6_x"
  bottom: "mean_scaled_flow_D_L6_x"
  top: "scaled_flow_D_nomean_L6_x"
  bias_param { axis: 0 }
}
layer {
  name: "mean_scaled_flow_D_L6_y"
  type: "Reduction"
  bottom: "scaled_flow_D_L6_y"
  top: "mean_scaled_flow_D_L6_y"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L6_y"
  type: "Bias"
  bottom: "scaled_flow_D_L6_y"
  bottom: "mean_scaled_flow_D_L6_y"
  top: "scaled_flow_D_nomean_L6_y"
  bias_param { axis: 0 }
}
layer {
  name: "FlowUnscaling_L6_R"
  type: "Eltwise"
  bottom: "scaled_flow_D2_L6"
  top: "flow_D2_L6"
  eltwise_param {
    operation: SUM
    coeff: 0.625
  }
}
layer {
  name: "coords_R_L6"
  type: "Eltwise"
  bottom: "flow_D2_L6"
  bottom: "gxy_L6"
  top: "coords_R_L6"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_img1_aug_L6"
  type: "Warp"
  bottom: "img1_aug_L6"
  bottom: "coords_R_L6"
  top: "warped_img1_aug_L6"
}
layer {
  name: "img_diff_L6"
  type: "Eltwise"
  bottom: "img0_aug_L6"
  bottom: "warped_img1_aug_L6"
  top: "img_diff_L6"
  eltwise_param {
    operation: SUM
    coeff: 1.0
    coeff: -1.0
  }
}
layer {
  name: "channelNorm_L6"
  type: "ChannelNorm"
  bottom: "img_diff_L6"
  top: "channelNorm_L6"
}
layer {
  name: "concat_F0_R_L6"
  type: "Concat"
  bottom: "channelNorm_L6"
  bottom: "scaled_flow_D_nomean_L6_x"
  bottom: "scaled_flow_D_nomean_L6_y"
  bottom: "F0_L6"
  top: "concat_F0_R_L6"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_R_L6"
  type: "Convolution"
  bottom: "concat_F0_R_L6"
  top: "conv1_R_L6"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv1_R_L6"
  type: "ReLU"
  bottom: "conv1_R_L6"
  top: "conv1_R_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_R_L6"
  type: "Convolution"
  bottom: "conv1_R_L6"
  top: "conv2_R_L6"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv2_R_L6"
  type: "ReLU"
  bottom: "conv2_R_L6"
  top: "conv2_R_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_R_L6"
  type: "Convolution"
  bottom: "conv2_R_L6"
  top: "conv3_R_L6"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv3_R_L6"
  type: "ReLU"
  bottom: "conv3_R_L6"
  top: "conv3_R_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_R_L6"
  type: "Convolution"
  bottom: "conv3_R_L6"
  top: "conv4_R_L6"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv4_R_L6"
  type: "ReLU"
  bottom: "conv4_R_L6"
  top: "conv4_R_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv5_R_L6"
  type: "Convolution"
  bottom: "conv4_R_L6"
  top: "conv5_R_L6"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv5_R_L6"
  type: "ReLU"
  bottom: "conv5_R_L6"
  top: "conv5_R_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv6_R_L6"
  type: "Convolution"
  bottom: "conv5_R_L6"
  top: "conv6_R_L6"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv6_R_L6"
  type: "ReLU"
  bottom: "conv6_R_L6"
  top: "conv6_R_L6"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "dist_R_L6"
  type: "Convolution"
  bottom: "conv6_R_L6"
  top: "dist_R_L6"
  convolution_param {
    num_output: 9
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "neg_sq_dist_R_L6"
  type: "NegSquare"
  bottom: "dist_R_L6"
  top: "neg_sq_dist_R_L6"
}
layer {
  name: "exp_kernel_R_L6"
  type: "ExpMax"
  bottom: "neg_sq_dist_R_L6"
  top: "exp_kernel_R_L6"
}
layer {
  name: "sum_exp_kernel_R_L6"
  type: "Convolution"
  bottom: "exp_kernel_R_L6"
  top: "sum_exp_kernel_R_L6"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "divisor_R_L6"
  bottom: "sum_exp_kernel_R_L6"
  top: "divisor_R_L6"
  type: "Power"
  power_param { power: -1 scale: 1 shift: 0 }
}
layer {
  name: "f-lconv_L6_x"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L6_x"
  bottom: "exp_kernel_R_L6"
  top: "f-lconv_L6_x"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L6_x"
  type: "Convolution"
  bottom: "f-lconv_L6_x"
  top: "scaled_flow0_R_L6_x"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L6_x"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L6_x"
  bottom: "divisor_R_L6"
  top: "scaled_flow1_R_L6_x"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "f-lconv_L6_y"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L6_y"
  bottom: "exp_kernel_R_L6"
  top: "f-lconv_L6_y"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L6_y"
  type: "Convolution"
  bottom: "f-lconv_L6_y"
  top: "scaled_flow0_R_L6_y"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L6_y"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L6_y"
  bottom: "divisor_R_L6"
  top: "scaled_flow1_R_L6_y"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "scaled_flow_R_L6"
  bottom: "scaled_flow1_R_L6_x"
  bottom: "scaled_flow1_R_L6_y"
  top: "scaled_flow_R_L6"
  type: "Concat"
  concat_param { axis: 1 }
}
#######################################
#            NetE-M: L5               #
#######################################
layer {
  name: "scaled_flow_R_L6to5"
  type: "Deconvolution"
  bottom: "scaled_flow_R_L6"
  top: "scaled_flow_R_L6to5"
  convolution_param {
    num_output: 2
    group: 2
    pad: 1
    kernel_size: 4
    stride: 2
    weight_filler { type: "bilinear" }
    bias_term: false
    engine: CUDNN
  }
}
layer {
  name: "FlowUnscaling_L5_D1"
  type: "Eltwise"
  bottom: "scaled_flow_R_L6to5"
  top: "flow_R_L6to5"
  eltwise_param {
    operation: SUM
    coeff: 1.25
  }
}
layer {
   name: "gxy_L5"
   type: "Grid"
   top: "gxy_L5"
   bottom: "flow_R_L6to5"
   propagate_down: false
}
layer {
  name: "coords_L5"
  type: "Eltwise"
  bottom: "flow_R_L6to5"
  bottom: "gxy_L5"
  top: "coords_L5"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_F1_L5"
  type: "Warp"
  bottom: "F1_L5"
  bottom: "coords_L5"
  top: "warped_F1_L5"
}
layer {
  name: "corr_L5"
  type: "Correlation"
  bottom: "F0_L5"
  bottom: "warped_F1_L5"
  top: "corr_L5"
  correlation_param {
    pad: 3
    kernel_size: 1
    max_displacement: 3
    stride_1: 1
    stride_2: 1
  }
}
layer {
  name: "ReLU_corr_L5"
  type: "ReLU"
  bottom: "corr_L5"
  top: "corr_L5"
  relu_param {
    negative_slope: 0.1
  }
}
layer {
  name: "conv1_D1_L5"
  type: "Convolution"
  bottom: "corr_L5"
  top: "conv1_D1_L5"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D1_L5"
  type: "ReLU"
  bottom: "conv1_D1_L5"
  top: "conv1_D1_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D1_L5"
  type: "Convolution"
  bottom: "conv1_D1_L5"
  top: "conv2_D1_L5"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D1_L5"
  type: "ReLU"
  bottom: "conv2_D1_L5"
  top: "conv2_D1_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D1_L5"
  type: "Convolution"
  bottom: "conv2_D1_L5"
  top: "conv3_D1_L5"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D1_L5"
  type: "ReLU"
  bottom: "conv3_D1_L5"
  top: "conv3_D1_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D1_res_L5"
  type: "Convolution"
  bottom: "conv3_D1_L5"
  top: "scaled_flow_D1_res_L5"
  convolution_param {
    num_output: 2
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D1_L5"
  type: "Eltwise"
  bottom: "scaled_flow_R_L6to5"
  bottom: "scaled_flow_D1_res_L5"
  top: "scaled_flow_D1_L5"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-S: L5              #
#######################################
layer {
  name: "FlowUnscaling_L5_D2"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L5"
  top: "flow_D1_L5"
  eltwise_param {
    operation: SUM
    coeff: 1.25
  }
}
layer {
  name: "coords_D1_L5"
  type: "Eltwise"
  bottom: "flow_D1_L5"
  bottom: "gxy_L5"
  top: "coords_D1_L5"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_D1_F1_L5"
  type: "Warp"
  bottom: "F1_L5"
  bottom: "coords_D1_L5"
  top: "warped_D1_F1_L5"
}
layer {
  name: "F_D2_L5"
  bottom: "F0_L5"
  bottom: "warped_D1_F1_L5"
  bottom: "scaled_flow_D1_L5"
  top: "F_D2_L5"
  type: "Concat"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_D2_L5"
  type: "Convolution"
  bottom: "F_D2_L5"
  top: "conv1_D2_L5"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D2_L5"
  type: "ReLU"
  bottom: "conv1_D2_L5"
  top: "conv1_D2_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D2_L5"
  type: "Convolution"
  bottom: "conv1_D2_L5"
  top: "conv2_D2_L5"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D2_L5"
  type: "ReLU"
  bottom: "conv2_D2_L5"
  top: "conv2_D2_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D2_L5"
  type: "Convolution"
  bottom: "conv2_D2_L5"
  top: "conv3_D2_L5"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D2_L5"
  type: "ReLU"
  bottom: "conv3_D2_L5"
  top: "conv3_D2_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D2_res_L5"
  type: "Convolution"
  bottom: "conv3_D2_L5"
  top: "scaled_flow_D2_res_L5"
  convolution_param {
    num_output: 2
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D2_L5"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L5"
  bottom: "scaled_flow_D2_res_L5"
  top: "scaled_flow_D2_L5"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-R: L5              #
#######################################
layer {
  name: "slice_scaled_flow_D_L5"
  type: "Slice"
  bottom: "scaled_flow_D2_L5"
  top: "scaled_flow_D_L5_x"
  top: "scaled_flow_D_L5_y"
  slice_param { axis: 1 slice_point: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L5_x"
  type: "Im2col"
  bottom: "scaled_flow_D_L5_x"
  top: "reshaped_scaled_flow_D_L5_x"
  convolution_param { pad: 1 kernel_size: 3 stride: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L5_y"
  type: "Im2col"
  bottom: "scaled_flow_D_L5_y"
  top: "reshaped_scaled_flow_D_L5_y"
  convolution_param { pad: 1 kernel_size: 3 stride: 1 }
}
layer {
  name: "mean_scaled_flow_D_L5_x"
  type: "Reduction"
  bottom: "scaled_flow_D_L5_x"
  top: "mean_scaled_flow_D_L5_x"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L5_x"
  type: "Bias"
  bottom: "scaled_flow_D_L5_x"
  bottom: "mean_scaled_flow_D_L5_x"
  top: "scaled_flow_D_nomean_L5_x"
  bias_param { axis: 0 }
}
layer {
  name: "mean_scaled_flow_D_L5_y"
  type: "Reduction"
  bottom: "scaled_flow_D_L5_y"
  top: "mean_scaled_flow_D_L5_y"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L5_y"
  type: "Bias"
  bottom: "scaled_flow_D_L5_y"
  bottom: "mean_scaled_flow_D_L5_y"
  top: "scaled_flow_D_nomean_L5_y"
  bias_param { axis: 0 }
}
layer {
  name: "FlowUnscaling_L5_R"
  type: "Eltwise"
  bottom: "scaled_flow_D2_L5"
  top: "flow_D2_L5"
  eltwise_param {
    operation: SUM
    coeff: 1.25
  }
}
layer {
  name: "coords_R_L5"
  type: "Eltwise"
  bottom: "flow_D2_L5"
  bottom: "gxy_L5"
  top: "coords_R_L5"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_img1_aug_L5"
  type: "Warp"
  bottom: "img1_aug_L5"
  bottom: "coords_R_L5"
  top: "warped_img1_aug_L5"
}
layer {
  name: "img_diff_L5"
  type: "Eltwise"
  bottom: "img0_aug_L5"
  bottom: "warped_img1_aug_L5"
  top: "img_diff_L5"
  eltwise_param {
    operation: SUM
    coeff: 1.0
    coeff: -1.0
  }
}
layer {
  name: "channelNorm_L5"
  type: "ChannelNorm"
  bottom: "img_diff_L5"
  top: "channelNorm_L5"
}
layer {
  name: "concat_F0_R_L5"
  type: "Concat"
  bottom: "channelNorm_L5"
  bottom: "scaled_flow_D_nomean_L5_x"
  bottom: "scaled_flow_D_nomean_L5_y"
  bottom: "F0_L5"
  top: "concat_F0_R_L5"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_R_L5"
  type: "Convolution"
  bottom: "concat_F0_R_L5"
  top: "conv1_R_L5"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv1_R_L5"
  type: "ReLU"
  bottom: "conv1_R_L5"
  top: "conv1_R_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_R_L5"
  type: "Convolution"
  bottom: "conv1_R_L5"
  top: "conv2_R_L5"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv2_R_L5"
  type: "ReLU"
  bottom: "conv2_R_L5"
  top: "conv2_R_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_R_L5"
  type: "Convolution"
  bottom: "conv2_R_L5"
  top: "conv3_R_L5"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv3_R_L5"
  type: "ReLU"
  bottom: "conv3_R_L5"
  top: "conv3_R_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_R_L5"
  type: "Convolution"
  bottom: "conv3_R_L5"
  top: "conv4_R_L5"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv4_R_L5"
  type: "ReLU"
  bottom: "conv4_R_L5"
  top: "conv4_R_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv5_R_L5"
  type: "Convolution"
  bottom: "conv4_R_L5"
  top: "conv5_R_L5"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv5_R_L5"
  type: "ReLU"
  bottom: "conv5_R_L5"
  top: "conv5_R_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv6_R_L5"
  type: "Convolution"
  bottom: "conv5_R_L5"
  top: "conv6_R_L5"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv6_R_L5"
  type: "ReLU"
  bottom: "conv6_R_L5"
  top: "conv6_R_L5"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "dist_R_L5"
  type: "Convolution"
  bottom: "conv6_R_L5"
  top: "dist_R_L5"
  convolution_param {
    num_output: 9
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "neg_sq_dist_R_L5"
  type: "NegSquare"
  bottom: "dist_R_L5"
  top: "neg_sq_dist_R_L5"
}
layer {
  name: "exp_kernel_R_L5"
  type: "ExpMax"
  bottom: "neg_sq_dist_R_L5"
  top: "exp_kernel_R_L5"
}
layer {
  name: "sum_exp_kernel_R_L5"
  type: "Convolution"
  bottom: "exp_kernel_R_L5"
  top: "sum_exp_kernel_R_L5"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "divisor_R_L5"
  bottom: "sum_exp_kernel_R_L5"
  top: "divisor_R_L5"
  type: "Power"
  power_param { power: -1 scale: 1 shift: 0 }
}
layer {
  name: "f-lconv_L5_x"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L5_x"
  bottom: "exp_kernel_R_L5"
  top: "f-lconv_L5_x"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L5_x"
  type: "Convolution"
  bottom: "f-lconv_L5_x"
  top: "scaled_flow0_R_L5_x"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L5_x"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L5_x"
  bottom: "divisor_R_L5"
  top: "scaled_flow1_R_L5_x"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "f-lconv_L5_y"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L5_y"
  bottom: "exp_kernel_R_L5"
  top: "f-lconv_L5_y"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L5_y"
  type: "Convolution"
  bottom: "f-lconv_L5_y"
  top: "scaled_flow0_R_L5_y"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L5_y"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L5_y"
  bottom: "divisor_R_L5"
  top: "scaled_flow1_R_L5_y"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "scaled_flow_R_L5"
  bottom: "scaled_flow1_R_L5_x"
  bottom: "scaled_flow1_R_L5_y"
  top: "scaled_flow_R_L5"
  type: "Concat"
  concat_param { axis: 1 }
}
#######################################
#             NetE-M: L4              #
#######################################
layer {
  name: "scaled_flow_R_L5to4"
  type: "Deconvolution"
  bottom: "scaled_flow_R_L5"
  top: "scaled_flow_R_L5to4"
  convolution_param {
    num_output: 2
    group: 2
    pad: 1
    kernel_size: 4
    stride: 2
    weight_filler { type: "bilinear" }
    bias_term: false
    engine: CUDNN
  }
}
layer {
  name: "FlowUnscaling_L4_D1"
  type: "Eltwise"
  bottom: "scaled_flow_R_L5to4"
  top: "flow_R_L5to4"
  eltwise_param {
    operation: SUM
    coeff: 2.5
  }
}
layer {
   name: "gxy_L4"
   type: "Grid"
   top: "gxy_L4"
   bottom: "flow_R_L5to4"
   propagate_down: false
}
layer {
  name: "coords_L4"
  type: "Eltwise"
  bottom: "flow_R_L5to4"
  bottom: "gxy_L4"
  top: "coords_L4"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_F1_L4"
  type: "Warp"
  bottom: "F1_L4"
  bottom: "coords_L4"
  top: "warped_F1_L4"
}
layer {
  name: "corr_L4"
  type: "Correlation"
  bottom: "F0_L4"
  bottom: "warped_F1_L4"
  top: "corr_L4"
  correlation_param {
    pad: 3
    kernel_size: 1
    max_displacement: 3
    stride_1: 1
    stride_2: 1
  }
}
layer {
  name: "ReLU_corr_L4"
  type: "ReLU"
  bottom: "corr_L4"
  top: "corr_L4"
  relu_param {
    negative_slope: 0.1
  }
}
layer {
  name: "conv1_D1_L4"
  type: "Convolution"
  bottom: "corr_L4"
  top: "conv1_D1_L4"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D1_L4"
  type: "ReLU"
  bottom: "conv1_D1_L4"
  top: "conv1_D1_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D1_L4"
  type: "Convolution"
  bottom: "conv1_D1_L4"
  top: "conv2_D1_L4"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D1_L4"
  type: "ReLU"
  bottom: "conv2_D1_L4"
  top: "conv2_D1_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D1_L4"
  type: "Convolution"
  bottom: "conv2_D1_L4"
  top: "conv3_D1_L4"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D1_L4"
  type: "ReLU"
  bottom: "conv3_D1_L4"
  top: "conv3_D1_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D1_res_L4"
  type: "Convolution"
  bottom: "conv3_D1_L4"
  top: "scaled_flow_D1_res_L4"
  convolution_param {
    num_output: 2
    pad: 2
    kernel_size: 5
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D1_L4"
  type: "Eltwise"
  bottom: "scaled_flow_R_L5to4"
  bottom: "scaled_flow_D1_res_L4"
  top: "scaled_flow_D1_L4"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-S: L4              #
#######################################
layer {
  name: "FlowUnscaling_L4_D2"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L4"
  top: "flow_D1_L4"
  eltwise_param {
    operation: SUM
    coeff: 2.5
  }
}
layer {
  name: "coords_D1_L4"
  type: "Eltwise"
  bottom: "flow_D1_L4"
  bottom: "gxy_L4"
  top: "coords_D1_L4"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_D1_F1_L4"
  type: "Warp"
  bottom: "F1_L4"
  bottom: "coords_D1_L4"
  top: "warped_D1_F1_L4"
}
layer {
  name: "F_D2_L4"
  bottom: "F0_L4"
  bottom: "warped_D1_F1_L4"
  bottom: "scaled_flow_D1_L4"
  top: "F_D2_L4"
  type: "Concat"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_D2_L4"
  type: "Convolution"
  bottom: "F_D2_L4"
  top: "conv1_D2_L4"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D2_L4"
  type: "ReLU"
  bottom: "conv1_D2_L4"
  top: "conv1_D2_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D2_L4"
  type: "Convolution"
  bottom: "conv1_D2_L4"
  top: "conv2_D2_L4"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D2_L4"
  type: "ReLU"
  bottom: "conv2_D2_L4"
  top: "conv2_D2_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D2_L4"
  type: "Convolution"
  bottom: "conv2_D2_L4"
  top: "conv3_D2_L4"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D2_L4"
  type: "ReLU"
  bottom: "conv3_D2_L4"
  top: "conv3_D2_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D2_res_L4"
  type: "Convolution"
  bottom: "conv3_D2_L4"
  top: "scaled_flow_D2_res_L4"
  convolution_param {
    num_output: 2
    pad: 2
    kernel_size: 5
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D2_L4"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L4"
  bottom: "scaled_flow_D2_res_L4"
  top: "scaled_flow_D2_L4"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-R: L4              #
#######################################
layer {
  name: "slice_scaled_flow_D_L4"
  type: "Slice"
  bottom: "scaled_flow_D2_L4"
  top: "scaled_flow_D_L4_x"
  top: "scaled_flow_D_L4_y"
  slice_param { axis: 1 slice_point: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L4_x"
  type: "Im2col"
  bottom: "scaled_flow_D_L4_x"
  top: "reshaped_scaled_flow_D_L4_x"
  convolution_param { pad: 2 kernel_size: 5 stride: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L4_y"
  type: "Im2col"
  bottom: "scaled_flow_D_L4_y"
  top: "reshaped_scaled_flow_D_L4_y"
  convolution_param { pad: 2 kernel_size: 5 stride: 1 }
}
layer {
  name: "mean_scaled_flow_D_L4_x"
  type: "Reduction"
  bottom: "scaled_flow_D_L4_x"
  top: "mean_scaled_flow_D_L4_x"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L4_x"
  type: "Bias"
  bottom: "scaled_flow_D_L4_x"
  bottom: "mean_scaled_flow_D_L4_x"
  top: "scaled_flow_D_nomean_L4_x"
  bias_param { axis: 0 }
}
layer {
  name: "mean_scaled_flow_D_L4_y"
  type: "Reduction"
  bottom: "scaled_flow_D_L4_y"
  top: "mean_scaled_flow_D_L4_y"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L4_y"
  type: "Bias"
  bottom: "scaled_flow_D_L4_y"
  bottom: "mean_scaled_flow_D_L4_y"
  top: "scaled_flow_D_nomean_L4_y"
  bias_param { axis: 0 }
}
layer {
  name: "FlowUnscaling_L4_R"
  type: "Eltwise"
  bottom: "scaled_flow_D2_L4"
  top: "flow_D2_L4"
  eltwise_param {
    operation: SUM
    coeff: 2.5
  }
}
layer {
  name: "coords_R_L4"
  type: "Eltwise"
  bottom: "flow_D2_L4"
  bottom: "gxy_L4"
  top: "coords_R_L4"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_img1_aug_L4"
  type: "Warp"
  bottom: "img1_aug_L4"
  bottom: "coords_R_L4"
  top: "warped_img1_aug_L4"
}
layer {
  name: "img_diff_L4"
  type: "Eltwise"
  bottom: "img0_aug_L4"
  bottom: "warped_img1_aug_L4"
  top: "img_diff_L4"
  eltwise_param {
    operation: SUM
    coeff: 1.0
    coeff: -1.0
  }
}
layer {
  name: "channelNorm_L4"
  type: "ChannelNorm"
  bottom: "img_diff_L4"
  top: "channelNorm_L4"
}
layer {
  name: "F0_128_L4"
  type: "Convolution"
  bottom: "F0_L4"
  top: "F0_128_L4"
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 1
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_F0_128_L4"
  type: "ReLU"
  bottom: "F0_128_L4"
  top: "F0_128_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "concat_F0_R_L4"
  type: "Concat"
  bottom: "channelNorm_L4"
  bottom: "scaled_flow_D_nomean_L4_x"
  bottom: "scaled_flow_D_nomean_L4_y"
  bottom: "F0_128_L4"
  top: "concat_F0_R_L4"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_R_L4"
  type: "Convolution"
  bottom: "concat_F0_R_L4"
  top: "conv1_R_L4"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv1_R_L4"
  type: "ReLU"
  bottom: "conv1_R_L4"
  top: "conv1_R_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_R_L4"
  type: "Convolution"
  bottom: "conv1_R_L4"
  top: "conv2_R_L4"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv2_R_L4"
  type: "ReLU"
  bottom: "conv2_R_L4"
  top: "conv2_R_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_R_L4"
  type: "Convolution"
  bottom: "conv2_R_L4"
  top: "conv3_R_L4"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv3_R_L4"
  type: "ReLU"
  bottom: "conv3_R_L4"
  top: "conv3_R_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_R_L4"
  type: "Convolution"
  bottom: "conv3_R_L4"
  top: "conv4_R_L4"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv4_R_L4"
  type: "ReLU"
  bottom: "conv4_R_L4"
  top: "conv4_R_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv5_R_L4"
  type: "Convolution"
  bottom: "conv4_R_L4"
  top: "conv5_R_L4"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv5_R_L4"
  type: "ReLU"
  bottom: "conv5_R_L4"
  top: "conv5_R_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv6_R_L4"
  type: "Convolution"
  bottom: "conv5_R_L4"
  top: "conv6_R_L4"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv6_R_L4"
  type: "ReLU"
  bottom: "conv6_R_L4"
  top: "conv6_R_L4"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "distH_R_L4"
  type: "Convolution"
  bottom: "conv6_R_L4"
  top: "distH_R_L4"
  convolution_param {
    num_output: 25
    pad_h: 2
    kernel_h: 5
    stride_h: 1
    pad_w: 0
    kernel_w: 1
    stride_w: 1
    engine: CUDNN
  }
}
layer {
  name: "distW_R_L4"
  type: "Convolution"
  bottom: "distH_R_L4"
  top: "distW_R_L4"
  convolution_param {
    num_output: 25
    pad_h: 0
    kernel_h: 1
    stride_h: 1
    pad_w: 2
    kernel_w: 5
    stride_w: 1
    engine: CUDNN
  }
}
layer {
  name: "neg_sq_dist_R_L4"
  type: "NegSquare"
  bottom: "distW_R_L4"
  top: "neg_sq_dist_R_L4"
}
layer {
  name: "exp_kernel_R_L4"
  type: "ExpMax"
  bottom: "neg_sq_dist_R_L4"
  top: "exp_kernel_R_L4"
}
layer {
  name: "sum_exp_kernel_R_L4"
  type: "Convolution"
  bottom: "exp_kernel_R_L4"
  top: "sum_exp_kernel_R_L4"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "divisor_R_L4"
  bottom: "sum_exp_kernel_R_L4"
  top: "divisor_R_L4"
  type: "Power"
  power_param { power: -1 scale: 1 shift: 0 }
}
layer {
  name: "f-lconv_L4_x"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L4_x"
  bottom: "exp_kernel_R_L4"
  top: "f-lconv_L4_x"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L4_x"
  type: "Convolution"
  bottom: "f-lconv_L4_x"
  top: "scaled_flow0_R_L4_x"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L4_x"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L4_x"
  bottom: "divisor_R_L4"
  top: "scaled_flow1_R_L4_x"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "f-lconv_L4_y"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L4_y"
  bottom: "exp_kernel_R_L4"
  top: "f-lconv_L4_y"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L4_y"
  type: "Convolution"
  bottom: "f-lconv_L4_y"
  top: "scaled_flow0_R_L4_y"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L4_y"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L4_y"
  bottom: "divisor_R_L4"
  top: "scaled_flow1_R_L4_y"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "scaled_flow_R_L4"
  bottom: "scaled_flow1_R_L4_x"
  bottom: "scaled_flow1_R_L4_y"
  top: "scaled_flow_R_L4"
  type: "Concat"
  concat_param { axis: 1 }
}
#######################################
#             NetE-M: L3              #
#######################################
layer {
  name: "scaled_flow_R_L4to3"
  type: "Deconvolution"
  bottom: "scaled_flow_R_L4"
  top: "scaled_flow_R_L4to3"
  convolution_param {
    num_output: 2
    group: 2
    pad: 1
    kernel_size: 4
    stride: 2
    weight_filler { type: "bilinear" }
    bias_term: false
    engine: CUDNN
  }
}
layer {
  name: "FlowUnscaling_L3_D1"
  type: "Eltwise"
  bottom: "scaled_flow_R_L4to3"
  top: "flow_R_L4to3"
  eltwise_param {
    operation: SUM
    coeff: 5.0
  }
}
layer {
   name: "gxy_L3"
   type: "Grid"
   top: "gxy_L3"
   bottom: "flow_R_L4to3"
   propagate_down: false
}
layer {
  name: "coords_L3"
  type: "Eltwise"
  bottom: "flow_R_L4to3"
  bottom: "gxy_L3"
  top: "coords_L3"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_F1_L3"
  type: "Warp"
  bottom: "F1_L3"
  bottom: "coords_L3"
  top: "warped_F1_L3"
}
layer {
  name: "corr_temp_L3"
  type: "Correlation"
  bottom: "F0_L3"
  bottom: "warped_F1_L3"
  top: "corr_temp_L3"
  correlation_param {
    pad: 6
    kernel_size: 1
    max_displacement: 6
    stride_1: 2
    stride_2: 2
  }
}
layer {
  name: "ReLU_corr_temp_L3"
  type: "ReLU"
  bottom: "corr_temp_L3"
  top: "corr_temp_L3"
  relu_param {
    negative_slope: 0.1
  }
}
layer {
  name: "corr_L3"
  type: "Deconvolution"
  bottom: "corr_temp_L3"
  top: "corr_L3"
  convolution_param {
    num_output: 49
    group: 49
    pad: 1
    kernel_size: 4
    stride: 2
    weight_filler { type: "bilinear" }
    bias_term: false
    engine: CUDNN
  }
}
layer {
  name: "conv1_D1_L3"
  type: "Convolution"
  bottom: "corr_L3"
  top: "conv1_D1_L3"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D1_L3"
  type: "ReLU"
  bottom: "conv1_D1_L3"
  top: "conv1_D1_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D1_L3"
  type: "Convolution"
  bottom: "conv1_D1_L3"
  top: "conv2_D1_L3"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D1_L3"
  type: "ReLU"
  bottom: "conv2_D1_L3"
  top: "conv2_D1_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D1_L3"
  type: "Convolution"
  bottom: "conv2_D1_L3"
  top: "conv3_D1_L3"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D1_L3"
  type: "ReLU"
  bottom: "conv3_D1_L3"
  top: "conv3_D1_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D1_res_L3"
  type: "Convolution"
  bottom: "conv3_D1_L3"
  top: "scaled_flow_D1_res_L3"
  convolution_param {
    num_output: 2
    pad: 2
    kernel_size: 5
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D1_L3"
  type: "Eltwise"
  bottom: "scaled_flow_R_L4to3"
  bottom: "scaled_flow_D1_res_L3"
  top: "scaled_flow_D1_L3"
  eltwise_param { operation: SUM }
}
#######################################
#            NetE-S: L3               #
#######################################
layer {
  name: "FlowUnscaling_L3_D2"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L3"
  top: "flow_D1_L3"
  eltwise_param {
    operation: SUM
    coeff: 5.0
  }
}
layer {
  name: "coords_D1_L3"
  type: "Eltwise"
  bottom: "flow_D1_L3"
  bottom: "gxy_L3"
  top: "coords_D1_L3"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_D1_F1_L3"
  type: "Warp"
  bottom: "F1_L3"
  bottom: "coords_D1_L3"
  top: "warped_D1_F1_L3"
}
layer {
  name: "F_D2_L3"
  bottom: "F0_L3"
  bottom: "warped_D1_F1_L3"
  bottom: "scaled_flow_D1_L3"
  top: "F_D2_L3"
  type: "Concat"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_D2_L3"
  type: "Convolution"
  bottom: "F_D2_L3"
  top: "conv1_D2_L3"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D2_L3"
  type: "ReLU"
  bottom: "conv1_D2_L3"
  top: "conv1_D2_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D2_L3"
  type: "Convolution"
  bottom: "conv1_D2_L3"
  top: "conv2_D2_L3"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D2_L3"
  type: "ReLU"
  bottom: "conv2_D2_L3"
  top: "conv2_D2_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D2_L3"
  type: "Convolution"
  bottom: "conv2_D2_L3"
  top: "conv3_D2_L3"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D2_L3"
  type: "ReLU"
  bottom: "conv3_D2_L3"
  top: "conv3_D2_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D2_res_L3"
  type: "Convolution"
  bottom: "conv3_D2_L3"
  top: "scaled_flow_D2_res_L3"
  convolution_param {
    num_output: 2
    pad: 2
    kernel_size: 5
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D2_L3"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L3"
  bottom: "scaled_flow_D2_res_L3"
  top: "scaled_flow_D2_L3"
  eltwise_param { operation: SUM }
}
#######################################
#            NetE-R: L3               #
#######################################
layer {
  name: "slice_scaled_flow_D_L3"
  type: "Slice"
  bottom: "scaled_flow_D2_L3"
  top: "scaled_flow_D_L3_x"
  top: "scaled_flow_D_L3_y"
  slice_param { axis: 1 slice_point: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L3_x"
  type: "Im2col"
  bottom: "scaled_flow_D_L3_x"
  top: "reshaped_scaled_flow_D_L3_x"
  convolution_param { pad: 2 kernel_size: 5 stride: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L3_y"
  type: "Im2col"
  bottom: "scaled_flow_D_L3_y"
  top: "reshaped_scaled_flow_D_L3_y"
  convolution_param { pad: 2 kernel_size: 5 stride: 1 }
}
layer {
  name: "mean_scaled_flow_D_L3_x"
  type: "Reduction"
  bottom: "scaled_flow_D_L3_x"
  top: "mean_scaled_flow_D_L3_x"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L3_x"
  type: "Bias"
  bottom: "scaled_flow_D_L3_x"
  bottom: "mean_scaled_flow_D_L3_x"
  top: "scaled_flow_D_nomean_L3_x"
  bias_param { axis: 0 }
}
layer {
  name: "mean_scaled_flow_D_L3_y"
  type: "Reduction"
  bottom: "scaled_flow_D_L3_y"
  top: "mean_scaled_flow_D_L3_y"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L3_y"
  type: "Bias"
  bottom: "scaled_flow_D_L3_y"
  bottom: "mean_scaled_flow_D_L3_y"
  top: "scaled_flow_D_nomean_L3_y"
  bias_param { axis: 0 }
}
layer {
  name: "FlowUnscaling_L3_R"
  type: "Eltwise"
  bottom: "scaled_flow_D2_L3"
  top: "flow_D2_L3"
  eltwise_param {
    operation: SUM
    coeff: 5.0
  }
}
layer {
  name: "coords_R_L3"
  type: "Eltwise"
  bottom: "flow_D2_L3"
  bottom: "gxy_L3"
  top: "coords_R_L3"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_img1_aug_L3"
  type: "Warp"
  bottom: "img1_aug_L3"
  bottom: "coords_R_L3"
  top: "warped_img1_aug_L3"
}
layer {
  name: "img_diff_L3"
  type: "Eltwise"
  bottom: "img0_aug_L3"
  bottom: "warped_img1_aug_L3"
  top: "img_diff_L3"
  eltwise_param {
    operation: SUM
    coeff: 1.0
    coeff: -1.0
  }
}
layer {
  name: "channelNorm_L3"
  type: "ChannelNorm"
  bottom: "img_diff_L3"
  top: "channelNorm_L3"
}
layer {
  name: "F0_128_L3"
  type: "Convolution"
  bottom: "F0_L3"
  top: "F0_128_L3"
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 1
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_F0_128_L3"
  type: "ReLU"
  bottom: "F0_128_L3"
  top: "F0_128_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "concat_F0_R_L3"
  type: "Concat"
  bottom: "channelNorm_L3"
  bottom: "scaled_flow_D_nomean_L3_x"
  bottom: "scaled_flow_D_nomean_L3_y"
  bottom: "F0_128_L3"
  top: "concat_F0_R_L3"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_R_L3"
  type: "Convolution"
  bottom: "concat_F0_R_L3"
  top: "conv1_R_L3"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv1_R_L3"
  type: "ReLU"
  bottom: "conv1_R_L3"
  top: "conv1_R_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_R_L3"
  type: "Convolution"
  bottom: "conv1_R_L3"
  top: "conv2_R_L3"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv2_R_L3"
  type: "ReLU"
  bottom: "conv2_R_L3"
  top: "conv2_R_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_R_L3"
  type: "Convolution"
  bottom: "conv2_R_L3"
  top: "conv3_R_L3"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv3_R_L3"
  type: "ReLU"
  bottom: "conv3_R_L3"
  top: "conv3_R_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_R_L3"
  type: "Convolution"
  bottom: "conv3_R_L3"
  top: "conv4_R_L3"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv4_R_L3"
  type: "ReLU"
  bottom: "conv4_R_L3"
  top: "conv4_R_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv5_R_L3"
  type: "Convolution"
  bottom: "conv4_R_L3"
  top: "conv5_R_L3"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv5_R_L3"
  type: "ReLU"
  bottom: "conv5_R_L3"
  top: "conv5_R_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv6_R_L3"
  type: "Convolution"
  bottom: "conv5_R_L3"
  top: "conv6_R_L3"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv6_R_L3"
  type: "ReLU"
  bottom: "conv6_R_L3"
  top: "conv6_R_L3"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "distH_R_L3"
  type: "Convolution"
  bottom: "conv6_R_L3"
  top: "distH_R_L3"
  convolution_param {
    num_output: 25
    pad_h: 2
    kernel_h: 5
    stride_h: 1
    pad_w: 0
    kernel_w: 1
    stride_w: 1
    engine: CUDNN
  }
}
layer {
  name: "distW_R_L3"
  type: "Convolution"
  bottom: "distH_R_L3"
  top: "distW_R_L3"
  convolution_param {
    num_output: 25
    pad_h: 0
    kernel_h: 1
    stride_h: 1
    pad_w: 2
    kernel_w: 5
    stride_w: 1
    engine: CUDNN
  }
}
layer {
  name: "neg_sq_dist_R_L3"
  type: "NegSquare"
  bottom: "distW_R_L3"
  top: "neg_sq_dist_R_L3"
}
layer {
  name: "exp_kernel_R_L3"
  type: "ExpMax"
  bottom: "neg_sq_dist_R_L3"
  top: "exp_kernel_R_L3"
}
layer {
  name: "sum_exp_kernel_R_L3"
  type: "Convolution"
  bottom: "exp_kernel_R_L3"
  top: "sum_exp_kernel_R_L3"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "divisor_R_L3"
  bottom: "sum_exp_kernel_R_L3"
  top: "divisor_R_L3"
  type: "Power"
  power_param { power: -1 scale: 1 shift: 0 }
}
layer {
  name: "f-lconv_L3_x"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L3_x"
  bottom: "exp_kernel_R_L3"
  top: "f-lconv_L3_x"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L3_x"
  type: "Convolution"
  bottom: "f-lconv_L3_x"
  top: "scaled_flow0_R_L3_x"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L3_x"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L3_x"
  bottom: "divisor_R_L3"
  top: "scaled_flow1_R_L3_x"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "f-lconv_L3_y"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L3_y"
  bottom: "exp_kernel_R_L3"
  top: "f-lconv_L3_y"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L3_y"
  type: "Convolution"
  bottom: "f-lconv_L3_y"
  top: "scaled_flow0_R_L3_y"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L3_y"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L3_y"
  bottom: "divisor_R_L3"
  top: "scaled_flow1_R_L3_y"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "scaled_flow_R_L3"
  bottom: "scaled_flow1_R_L3_x"
  bottom: "scaled_flow1_R_L3_y"
  top: "scaled_flow_R_L3"
  type: "Concat"
  concat_param { axis: 1 }
}
#######################################
#             NetE-M: L2              #
#######################################
layer {
  name: "F0_F1_64_L2"
  type: "Convolution"
  bottom: "F0_L2"
  bottom: "F1_L2"
  top: "F0_64_L2"
  top: "F1_64_L2"
  convolution_param {
    num_output: 64
    pad: 0
    kernel_size: 1
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_F0_64_L2"
  type: "ReLU"
  bottom: "F0_64_L2"
  top: "F0_64_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "relu_F1_64_L2"
  type: "ReLU"
  bottom: "F1_64_L2"
  top: "F1_64_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_R_L3to2"
  type: "Deconvolution"
  bottom: "scaled_flow_R_L3"
  top: "scaled_flow_R_L3to2"
  convolution_param {
    num_output: 2
    group: 2
    pad: 1
    kernel_size: 4
    stride: 2
    weight_filler { type: "bilinear" }
    bias_term: false
    engine: CUDNN
  }
}
layer {
  name: "FlowUnscaling_L2_D1"
  type: "Eltwise"
  bottom: "scaled_flow_R_L3to2"
  top: "flow_R_L3to2"
  eltwise_param {
    operation: SUM
    coeff: 10.0
  }
}
layer {
   name: "gxy_L2"
   type: "Grid"
   top: "gxy_L2"
   bottom: "flow_R_L3to2"
   propagate_down: false
}
layer {
  name: "coords_L2"
  type: "Eltwise"
  bottom: "flow_R_L3to2"
  bottom: "gxy_L2"
  top: "coords_L2"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_F1_64_L2"
  type: "Warp"
  bottom: "F1_64_L2"
  bottom: "coords_L2"
  top: "warped_F1_64_L2"
}
layer {
  name: "corr_temp_L2"
  type: "Correlation"
  bottom: "F0_64_L2"
  bottom: "warped_F1_64_L2"
  top: "corr_temp_L2"
  correlation_param {
    pad: 6
    kernel_size: 1
    max_displacement: 6
    stride_1: 2
    stride_2: 2
  }
}
layer {
  name: "ReLU_corr_temp_L2"
  type: "ReLU"
  bottom: "corr_temp_L2"
  top: "corr_temp_L2"
  relu_param {
    negative_slope: 0.1
  }
}
layer {
  name: "corr_L2"
  type: "Deconvolution"
  bottom: "corr_temp_L2"
  top: "corr_L2"
  convolution_param {
    num_output: 49
    group: 49
    pad: 1
    kernel_size: 4
    stride: 2
    weight_filler { type: "bilinear" }
    bias_term: false
    engine: CUDNN
  }
}
layer {
  name: "conv1_D1_L2"
  type: "Convolution"
  bottom: "corr_L2"
  top: "conv1_D1_L2"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D1_L2"
  type: "ReLU"
  bottom: "conv1_D1_L2"
  top: "conv1_D1_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D1_L2"
  type: "Convolution"
  bottom: "conv1_D1_L2"
  top: "conv2_D1_L2"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D1_L2"
  type: "ReLU"
  bottom: "conv2_D1_L2"
  top: "conv2_D1_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D1_L2"
  type: "Convolution"
  bottom: "conv2_D1_L2"
  top: "conv3_D1_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D1_L2"
  type: "ReLU"
  bottom: "conv3_D1_L2"
  top: "conv3_D1_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D1_res_L2"
  type: "Convolution"
  bottom: "conv3_D1_L2"
  top: "scaled_flow_D1_res_L2"
  convolution_param {
    num_output: 2
    pad: 3
    kernel_size: 7
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D1_L2"
  type: "Eltwise"
  bottom: "scaled_flow_R_L3to2"
  bottom: "scaled_flow_D1_res_L2"
  top: "scaled_flow_D1_L2"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-S: L2              #
#######################################
layer {
  name: "FlowUnscaling_L2_D2"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L2"
  top: "flow_D1_L2"
  eltwise_param {
    operation: SUM
    coeff: 10.0
  }
}
layer {
  name: "coords_D1_L2"
  type: "Eltwise"
  bottom: "flow_D1_L2"
  bottom: "gxy_L2"
  top: "coords_D1_L2"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_D1_F1_64_L2"
  type: "Warp"
  bottom: "F1_64_L2"
  bottom: "coords_D1_L2"
  top: "warped_D1_F1_64_L2"
}
layer {
  name: "F_D2_L2"
  bottom: "F0_64_L2"
  bottom: "warped_D1_F1_64_L2"
  bottom: "scaled_flow_D1_L2"
  top: "F_D2_L2"
  type: "Concat"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_D2_L2"
  type: "Convolution"
  bottom: "F_D2_L2"
  top: "conv1_D2_L2"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU1_D2_L2"
  type: "ReLU"
  bottom: "conv1_D2_L2"
  top: "conv1_D2_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_D2_L2"
  type: "Convolution"
  bottom: "conv1_D2_L2"
  top: "conv2_D2_L2"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU2_D2_L2"
  type: "ReLU"
  bottom: "conv2_D2_L2"
  top: "conv2_D2_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_D2_L2"
  type: "Convolution"
  bottom: "conv2_D2_L2"
  top: "conv3_D2_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "ReLU3_D2_L2"
  type: "ReLU"
  bottom: "conv3_D2_L2"
  top: "conv3_D2_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "scaled_flow_D2_res_L2"
  type: "Convolution"
  bottom: "conv3_D2_L2"
  top: "scaled_flow_D2_res_L2"
  convolution_param {
    num_output: 2
    pad: 3
    kernel_size: 7
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "scaled_flow_D2_L2"
  type: "Eltwise"
  bottom: "scaled_flow_D1_L2"
  bottom: "scaled_flow_D2_res_L2"
  top: "scaled_flow_D2_L2"
  eltwise_param { operation: SUM }
}
#######################################
#             NetE-R: L2              #
#######################################
layer {
  name: "slice_scaled_flow_D_L2"
  type: "Slice"
  bottom: "scaled_flow_D2_L2"
  top: "scaled_flow_D_L2_x"
  top: "scaled_flow_D_L2_y"
  slice_param { axis: 1 slice_point: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L2_x"
  type: "Im2col"
  bottom: "scaled_flow_D_L2_x"
  top: "reshaped_scaled_flow_D_L2_x"
  convolution_param { pad: 3 kernel_size: 7 stride: 1 }
}
layer {
  name: "reshaped_scaled_flow_D_L2_y"
  type: "Im2col"
  bottom: "scaled_flow_D_L2_y"
  top: "reshaped_scaled_flow_D_L2_y"
  convolution_param { pad: 3 kernel_size: 7 stride: 1 }
}
layer {
  name: "mean_scaled_flow_D_L2_x"
  type: "Reduction"
  bottom: "scaled_flow_D_L2_x"
  top: "mean_scaled_flow_D_L2_x"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L2_x"
  type: "Bias"
  bottom: "scaled_flow_D_L2_x"
  bottom: "mean_scaled_flow_D_L2_x"
  top: "scaled_flow_D_nomean_L2_x"
  bias_param { axis: 0 }
}
layer {
  name: "mean_scaled_flow_D_L2_y"
  type: "Reduction"
  bottom: "scaled_flow_D_L2_y"
  top: "mean_scaled_flow_D_L2_y"
  reduction_param { operation: MEAN axis: 1 coeff: -1 }
}
layer {
  name: "scaled_flow_D_nomean_L2_y"
  type: "Bias"
  bottom: "scaled_flow_D_L2_y"
  bottom: "mean_scaled_flow_D_L2_y"
  top: "scaled_flow_D_nomean_L2_y"
  bias_param { axis: 0 }
}
layer {
  name: "FlowUnscaling_L2_R"
  type: "Eltwise"
  bottom: "scaled_flow_D2_L2"
  top: "flow_D2_L2"
  eltwise_param {
    operation: SUM
    coeff: 10.0
  }
}
layer {
  name: "coords_R_L2"
  type: "Eltwise"
  bottom: "flow_D2_L2"
  bottom: "gxy_L2"
  top: "coords_R_L2"
  eltwise_param { coeff: 1 coeff: 1 }
}
layer {
  name: "warped_img1_aug_L2"
  type: "Warp"
  bottom: "img1_aug_L2"
  bottom: "coords_R_L2"
  top: "warped_img1_aug_L2"
}
layer {
  name: "img_diff_L2"
  type: "Eltwise"
  bottom: "img0_aug_L2"
  bottom: "warped_img1_aug_L2"
  top: "img_diff_L2"
  eltwise_param {
    operation: SUM
    coeff: 1.0
    coeff: -1.0
  }
}
layer {
  name: "channelNorm_L2"
  type: "ChannelNorm"
  bottom: "img_diff_L2"
  top: "channelNorm_L2"
}
layer {
  name: "F0_128_L2"
  type: "Convolution"
  bottom: "F0_L2"
  top: "F0_128_L2"
  convolution_param {
    num_output: 128
    pad: 0
    kernel_size: 1
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_F0_128_L2"
  type: "ReLU"
  bottom: "F0_128_L2"
  top: "F0_128_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "concat_F0_R_L2"
  type: "Concat"
  bottom: "channelNorm_L2"
  bottom: "scaled_flow_D_nomean_L2_x"
  bottom: "scaled_flow_D_nomean_L2_y"
  bottom: "F0_128_L2"
  top: "concat_F0_R_L2"
  concat_param { axis: 1 }
}
layer {
  name: "conv1_R_L2"
  type: "Convolution"
  bottom: "concat_F0_R_L2"
  top: "conv1_R_L2"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv1_R_L2"
  type: "ReLU"
  bottom: "conv1_R_L2"
  top: "conv1_R_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv2_R_L2"
  type: "Convolution"
  bottom: "conv1_R_L2"
  top: "conv2_R_L2"
  convolution_param {
    num_output: 128
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv2_R_L2"
  type: "ReLU"
  bottom: "conv2_R_L2"
  top: "conv2_R_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv3_R_L2"
  type: "Convolution"
  bottom: "conv2_R_L2"
  top: "conv3_R_L2"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv3_R_L2"
  type: "ReLU"
  bottom: "conv3_R_L2"
  top: "conv3_R_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv4_R_L2"
  type: "Convolution"
  bottom: "conv3_R_L2"
  top: "conv4_R_L2"
  convolution_param {
    num_output: 64
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv4_R_L2"
  type: "ReLU"
  bottom: "conv4_R_L2"
  top: "conv4_R_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv5_R_L2"
  type: "Convolution"
  bottom: "conv4_R_L2"
  top: "conv5_R_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv5_R_L2"
  type: "ReLU"
  bottom: "conv5_R_L2"
  top: "conv5_R_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "conv6_R_L2"
  type: "Convolution"
  bottom: "conv5_R_L2"
  top: "conv6_R_L2"
  convolution_param {
    num_output: 32
    pad: 1
    kernel_size: 3
    stride: 1
    engine: CUDNN
  }
}
layer {
  name: "relu_conv6_R_L2"
  type: "ReLU"
  bottom: "conv6_R_L2"
  top: "conv6_R_L2"
  relu_param { negative_slope: 0.1 }
}
layer {
  name: "distH_R_L2"
  type: "Convolution"
  bottom: "conv6_R_L2"
  top: "distH_R_L2"
  convolution_param {
    num_output: 49
    pad_h: 3
    kernel_h: 7
    stride_h: 1
    pad_w: 0
    kernel_w: 1
    stride_w: 1
    engine: CUDNN
  }
}
layer {
  name: "distW_R_L2"
  type: "Convolution"
  bottom: "distH_R_L2"
  top: "distW_R_L2"
  convolution_param {
    num_output: 49
    pad_h: 0
    kernel_h: 1
    stride_h: 1
    pad_w: 3
    kernel_w: 7
    stride_w: 1
    engine: CUDNN
  }
}
layer {
  name: "neg_sq_dist_R_L2"
  type: "NegSquare"
  bottom: "distW_R_L2"
  top: "neg_sq_dist_R_L2"
}
layer {
  name: "exp_kernel_R_L2"
  type: "ExpMax"
  bottom: "neg_sq_dist_R_L2"
  top: "exp_kernel_R_L2"
}
layer {
  name: "sum_exp_kernel_R_L2"
  type: "Convolution"
  bottom: "exp_kernel_R_L2"
  top: "sum_exp_kernel_R_L2"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "divisor_R_L2"
  bottom: "sum_exp_kernel_R_L2"
  top: "divisor_R_L2"
  type: "Power"
  power_param { power: -1 scale: 1 shift: 0 }
}
layer {
  name: "f-lconv_L2_x"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L2_x"
  bottom: "exp_kernel_R_L2"
  top: "f-lconv_L2_x"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L2_x"
  type: "Convolution"
  bottom: "f-lconv_L2_x"
  top: "scaled_flow0_R_L2_x"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L2_x"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L2_x"
  bottom: "divisor_R_L2"
  top: "scaled_flow1_R_L2_x"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "f-lconv_L2_y"
  type: "Eltwise"
  bottom: "reshaped_scaled_flow_D_L2_y"
  bottom: "exp_kernel_R_L2"
  top: "f-lconv_L2_y"
  eltwise_param { operation: PROD }
}
layer {
  name: "scaled_flow0_R_L2_y"
  type: "Convolution"
  bottom: "f-lconv_L2_y"
  top: "scaled_flow0_R_L2_y"
  convolution_param {
    num_output: 1
    kernel_size: 1
    weight_filler { type: "constant" value: 1 }
  }
}
layer {
  name: "scaled_flow1_R_L2_y"
  type: "Eltwise"
  bottom: "scaled_flow0_R_L2_y"
  bottom: "divisor_R_L2"
  top: "scaled_flow1_R_L2_y"
  eltwise_param {
     operation: PROD
  }
}
layer {
  name: "scaled_flow_R_L2"
  bottom: "scaled_flow1_R_L2_x"
  bottom: "scaled_flow1_R_L2_y"
  top: "scaled_flow_R_L2"
  type: "Concat"
  concat_param { axis: 1 }
}
#######################################
#           Post-processing           #
#######################################
layer {
  name: "FlowUnscaling"
  type: "Eltwise"
  bottom: "scaled_flow_R_L2"
  top: "resized_flow"
  eltwise_param { operation: SUM coeff: 20.0 }
  propagate_down: false
}
layer {
  name: "Resample_flow"
  type: "Resample"
  bottom: "resized_flow"
  top: "flow"
  resample_param {
    width: 320
    height: 240
    type: LINEAR
    antialias: true
  }
}
layer {
  name: "ScaleMag_flow"
  type: "Convolution"
  bottom: "flow"
  top: "final_flow"
  convolution_param {
    num_output: 2
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "diagonal"
      diag_val: 1.00000000
      diag_val: 0.93750000
    }
    bias_filler {
      type: "constant"
    }
  }
}