[ResNet18 DepthNet]
  [Conv2d]
    * K: kernel size = 3
    * S: stride = 1

  [Max Pooling]
    * K: kernel size = 2

  [ResidualBlock]
    #1 Conv2d + BatchNorm + ReLU
    #2 Conv2d + BatchNorm
    #3 Concat (#1, #2, D=1)

  #0 Input RGB image = 3, H, W

  #1 [Depth Encoder]
    #1 Conv2d (K=7, S=2) + BatchNorm + ReLU = 64, H/2, W/2
    #2 Conv2d + BatchNorm + ReLU = 64, H/2, W/2
    #3 ResidualBlock (#2) *2 = 64, H/2, W/2
    #4 Max Pooling = 64, H/4, W/4
    #5 ResidualBlock (#3 + #2) *2 = 128, H/4, W/4