Reducing stereo ray-tracing budget with 3D warping and perception-driven reconstruction

Abstract

Monte Carlo ray tracing can produce high-fidelity stereo renderings of 3D scenes. However, the computational costs associated with shading indirect illumination can be prohibitively expensive. While techniques like reprojection and reconstruction have been employed to reduce the inherent redundancy in stereo rendering, they often fall short in addressing issues like remaining holes and appearing artifacts, particularly in the case of specular reflections. In this paper, we introduce a stereo rendering method that leverages both ray tracing and 3D warping, incorporating techniques for reusing indirect illumination calculations and perception-driven reconstructions. At the pre-rendering stage, we reuse indirect illumination calculations in the background and 3D warping regions, resulting in significant savings of indirect illumination calculations in the testing scenes. For the remaining holes, we adopt a sparse sampling strategy for representative pixels, guided by saliency maps. We iteratively reconstruct indirect illumination, employing a stopping criterion based on just noticeable difference thresholds. Through equal-sample experiments, we showcase the efficiency achieved by the proposed method in reducing high-level sampling costs for 87.688% ~ 99.581% regions of the test scenes. Furthermore, the proposed method excels in accurately rendering both diffuse and specular reflections, ultimately enhancing the overall visual perception quality.