Image-space Control Variates for Rendering

摘要

We explore the theory of integration with control variates in therncontext of rendering. Our goal is to optimally combine multiplernestimators using their covariances. We focus on two applications,rnre-rendering and gradient-domain rendering, where we exploit coherencernbetween temporally and spatially adjacent pixels. We proposernan image-space (iterative) reconstruction scheme that employsrncontrol variates to reduce variance. We show that recent works onrnscene editing and gradient-domain rendering can be directly formulatedrnas control-variate estimators, despite using seemingly differentrnapproaches. In particular, we demonstrate the conceptual equivalencernof screened Poisson image reconstruction and our iterativernreconstruction scheme. Our composite estimators offer practicalrnand simple solutions that improve upon the current state of the artrnfor the two investigated applications.