Progressive volume photon tracing

Over the past decade, photon mapping is being more and more used to simulate light transport in scenes containing participating media. However, photon mapping is limited practically because of its large memory requirements. Progressive photon mapping addresses this problem [Hachisuka et al. 2008; Knaus and Zwicker 2011]. This approach computes many images of the same scene from the same viewpoint using different sets of photons and different parameter values. The obtained low quality images, when combined together, generate images of higher quality. In this talk, we propose a volume photon mapping method which does not have to store the photon maps, ensures a faster convergence to the final solution, and allows the user to have a full control of the size of the used memory. We assign a beam to each primary or secondary ray crossing a participating medium, then build a beam hierarchy. As soon as the computation of an emmited photon's contribution to a beam or to a visible point is done, we throw away this photon. In this way there is no need to store any photon map, which solves the problem of memory storage.