Test Suite Overview


We present equal-time and equal-variance comparisons between brute-force reference techniques, the method of Meng et al. [2015], and ours. Due to the way we render images on a cluster we can not show images with exactly equal render times and variance. Thus, we always give our methods the disadvantage by displaying images with lowest render times in equal-time comparisons and lowest variances in equal-variance comparisons.

For each rendering technique we report time to unit variance (TTUV), the mean relative squared error (MRSE), and an estimate of the absolute bias. We compute the latter by subtracting a variance estimate of both images from the mean squared error (MSE) and then taking the square root.


Bowl

Castle

Dunes

Lucy

Pile

TwoPiles

Snowman

Spices

Wedge