LinkNet (Dataset: CamVid)

Here the network is being run on Titan X GPU. CamVid dataset has images of resolution 960x720, but for this experiment they were downsampled by a factor of 1.25. fps displayed in the bottom corner involves three timing values:

  • Loading and downsampling of input image
  • Feed forward time of network
  • Overlaying the output on input image and displaying them