High Efficiency Real Time Rendering for Point-Based Model on GPU

In the last years the increasing programmability of GPU lead to lots of point-based rendering could be implemented on GPU, which not only let CPU to be available for other tasks, but also improve the execution speed through GPU. However, since the recent GPU does not support the epsilon-z-buffering algorithm, a majority of GPU based rendering algorithms are based on multi-pass rendering to achieve high visual quality. These approaches need to rasterize lots of possibly visible splats in the first and second pass, and implement lots of computations on these possibly visible splats in the pixel shader in the second pass. In this paper we propose an improved algorithm of multi-pass rendering based on modern GPU. Comparing with the existing multi-pass methods, our approach could acquire all visible splats after rasterization and depth test in the first pass. Then, we only rasterize and per-pixel compute these visible splats in the second pass, which effectively avoid unnecessary shading computations. Experiment results show that this algorithm could greatly improve the rendering efficiency and achieve the visual quality comparable to the previous algorithms.