Uma arquitetura paralela para a renderização de múltiplos pontos de vista

Abstract

We present a parallel architecture to e_ciently render multiple viewpoints on a cluster of GPUs. Our approach employs a light _eld renderer to reconstruct the desired views from a set of images of the scene. To allow the rendering of dynamic datasets, we sample the models for every frame, instead of sampling once and storing the resulting images in a bu_er. At the same time, our parallelization allows each view to use any of the generated samples, avoiding redundant work. The e_cieny of the solution was inspected through mathematical models and validated by measurements of speedup and e_ciency. We concluded that the proposed solution is scalable and can support rendering at interactive rates.