Progressive photon mapping was first proposed by Hachisuka et al. (2008) as an iterative extension of the standard static photon mapping approach as implemented in the Radiance extension. It combines multiple smaller photon maps to approximate a much larger one which may not fit into memory using the traditional approach. Through iteration, the process mitigates the noise inherent in Monte Carlo raytracing by combining successive results and averaging them. At the same time, the density estimate bandwidth1Bandwidth describes the support, or area of influence, of a filter used to weight the photons retrieved from the photon map during a nearest neighbour lookup on a surface (Jensen, 2001). The resulting irradiance is proportional to the photon density, and the bandwidth is defined by the distance (radius) to the furthest photon found. In this paper, we generalise the term to describe either the radius or the number of nearest neighbours for a density estimate, depending on the implementation.1 (radius or number of nearest photons) is gradually reduced to mitigate bias. As Hachisuka points out, the accumulated density estimates converge to an unbiased solution in the limit.
