Since rendering is a stochastic process, it would be useful to compare repeated renderings by the same renderer, and to assess the variance of the renderings.
Many factors would affect rendering variance, including choice of integrator, sampling strategy, and number of samples. This would potentially be a huge space to explore.
Realistically, we could pick one existing example scene and render it twice using "normal" factors like the path tracing integrator, low discrepancy sampler, and enough samples that the image "looks good" and contains no obvious artifacts. TableSphere, SpectralIllusion, and Interior are all reasonable candidates.
We could implement this comparison as a scene variant: it would use the existing parent scene but make a new conditions file that specified repeated rendering (with different imageNames). A "make figure" scrip could plot the repeated renderings and a difference image, and maybe compare image "slices".
Since rendering is a stochastic process, it would be useful to compare repeated renderings by the same renderer, and to assess the variance of the renderings.
Many factors would affect rendering variance, including choice of integrator, sampling strategy, and number of samples. This would potentially be a huge space to explore.
Realistically, we could pick one existing example scene and render it twice using "normal" factors like the path tracing integrator, low discrepancy sampler, and enough samples that the image "looks good" and contains no obvious artifacts. TableSphere, SpectralIllusion, and Interior are all reasonable candidates.
We could implement this comparison as a scene variant: it would use the existing parent scene but make a new conditions file that specified repeated rendering (with different imageNames). A "make figure" scrip could plot the repeated renderings and a difference image, and maybe compare image "slices".