Efficient Stereo-Aware Screen-Space Ambient Occlusion with Adaptive Computation
Yu-Ting WuNational Taipei University
Visualization of the stereo-inconsistent pixels. The leftmost image displays a view of the SPONZA scene, while the second to fourth images show the ambient occlusion (AO) of that view, generated by various methods using equal samples. Blue pixels represent surfaces visible to only one view. Red pixels highlight stereo-inconsistent areas where differences between views exceed 2%, which could be noticeable to humans according to Weber’s law. Applying the monoscopic SSAO method directly in stereo rendering results in numerous inconsistent pixels (7.09%), potentially disrupting depth perception. Shi et al. mitigate these inconsistencies by applying stereo-aware computations to every pixel, though this comes with performance drawbacks. Our method delivers comparable results with significantly improved efficiency. The execution time includes obscurance estimation and cross-bilateral filtering. Due to space constraints, only the right view is shown here.
Abstract
Screen-space ambient occlusion (SSAO) has become a widely used technique in real-time rendering, valued for its high performance and full support for dynamic geometry. However, applying SSAO directly to stereo rendering can result in incorrect depth perception and viewer discomfort due to differences in captured scene approximations between the left and right views. Existing methods for generating stereo-consistent SSAO often involve substantial computational costs. This paper introduces an adaptive method, inspired by Weber’s law, to enhance the efficiency of achieving stereo-consistent SSAO. Our method identifies inconsistent pixels generated by cost-effective SSAO algorithms, such as monoscopic SSAO, and selectively applies computationally intensive stereo-aware computations only to those pixels. Experiments demonstrate that our method delivers stereo-consistent results comparable to state-of-the-art techniques while significantly enhancing rendering performance.
Publication
Yu-Ting Wu.
Efficient Stereo-Aware Screen-Space Ambient Occlusion with Adaptive Computation.
IEEE Computer Graphics and Applications (CG&A). BibTeX (coming soon)
IEEE CG&A 2025 paper (coming soon)
Digital library (coming soon)
Supplemental
Additional results and comparisons (coming soon)