Examensarbetspresentation Alexander Hansson och Peter Moodie
Image quality metrics are used to evaluate the percieved quality of processed images.
Differences in hardware between graphics processors contribute to noise during quality eval-
uation. In this masters thesis paper we train and evaluate neural networks as metrics to
evaluate GPU rendering quality. The neural networks can successfully ignore the rendering
noise that occurs when the test and reference frames are rendered by different GPUs. This
reduces tedious human interaction which requires manual updates of reference-frames during
Magnus Oskarsson, Centre for Mathematical Sciences, Lund University Examiner Mikael Nilsson, Centre for Mathematical Sciences, Lund University