Spatiotemporal Resampling-Based Real-Time Stochastic Lightcuts

This paper presented a spatiotemporal resampling algorithm for rendering dynamic scenes with many dynamic lights. Firstly, performed importance sampling by real-time stochastic lightcuts to generate resampling samples. Secondly, a fast algorithm to judge the validity of spatiotemporal samples is proposed, which avoided errors due to wrong selection of resampling samples by checking geometric properties and visibility. Finally, resampled spatiotemporal samples and selected important samples by weight to improve quality of samples. The experimental results in self-building scenes and NVIDIA ORCA indicate that this algorithm can achieve good sampling quality, and the time consuming and rendering quality of this algorithm can be balanced by modifying the spatial resampling quantity to make it scalable. Compared with real-time stochastic lightcuts algorithm, this algorithm can achieve higher rendering quality with same sample quantity and reduces 20% to 40% sampling consumption with similar rendering quality.