No abstract

Dragon sequence on light eld display VPT -64 spp -75 min MVPT -16 spp natives -75 min VPT SMAPE 0.021 relMSE 0.003 RMSE 0.029 MVPT SMAPE 0.011 relMSE 0.001 RMSE 0.016Figure 1:We render a sequence of 45 frames of the Dragon cloud scene including heterogeneous medium to be displayed on a lightfield screen. Instead of rendering the sequence frame by frame using volumetric path tracing (VPT), our method (MVPT) jointly renders the sequence at once reusing some computation during the simulation. We observe a significant variance reduction at equal time when rendering the whole sequence -or reach a desired quality at lower time. In addition the correlation of reused paths between nearby frames almost makes the flickering caused by dissimilar random sampling dissapear. This makes the sequences pleasier to the eye even at low sample count.