Population Monte Carlo Samplers for Rendering

Abstract

We present novel samplers and algorithms for Monte Carlo rendering. The adaptive image-plane sampler selects pixels for refinement according to a perceptually-weighted variance criteria. The hemispheric integrals sampler learns an importance sampling function for computing common rendering integrals. Both algorithms, which are unbiased, are derived in the generic Population Monte Carlo statistical framework, which works on a population of samples that is iterated through distributions that are modified over time. Information found in one iteration can be used to guide subsequent iterations. Our results improve rendering efficiency by a factor of between 2 to 5 over existing techniques. We also show how both samplers can be easily incorporated into a global rendering system.