A memory-efficient scheme for fast spectral photon mapping

We propose a new data structure for spectral rendering using photon mapping. Although spectral rendering can generate high quality realistic images with photon mapping, naive combination of spectral rendering and photon mapping leads to deterioration in both performance and memory usage. The proposed data structure creates a coherent access pattern to the main memory and achieves acceleration of rendering. The data structure also remarkably saves memory space with a low overhead. It can be applicable to many other techniques proposed so far such as volume photon mapping with spectral effects, because of its simplicity.