high performance computing on graphics processing units: hgpu.org

We describe a generic approach to incorporate progressive refinement into GPU-based ray casting. Our approach allows to interactively navigate through highly complex scenes that may usually take several seconds to render while producing high-quality anti-aliased images in late stages of the refinement process. It maintains interactivity by initially evaluating only a small number of screen space samples (i.e., casting only few rays) and generating a possibly blurry initial frame. As long as the user does not interact with the scene, the approach proceeds to evaluate further batches of samples resulting in increasingly sharp intermediate frames. This smooth progression continues until a final anti-aliased image is obtained. We use a hierarchical Poisson disk sampling pattern to achieve progressive sampling as well as blue noise sampling and supersampling. To reconstruct intermediate frames, we use an adaptive weighted average filtering scheme. Our approach solely relies on the evaluation of samples in screen space, the actual ray casting algorithm is treated as an interchangeable black box. Therefore, the approach is suitable for a wide range of applications. Our implementation as part of a volume ray casting framework demonstrates how the approach can exploit modern graphics hardware.