Frame-to-frame coherent GPU splatting

Abstract

Recently, several improvements of the classical splatting volume rendering technique have been proposed, specifically, GPU implementations that considerably speed it up. However, splatting time-varying data, even using the GPU, is still slow and not very suitable for interactive data exploration through time. In this paper, we propose a GPU-based viewaligned splatting algorithm that exploits frame-to-frame coherence to render time-varying volume datasets. Our method run-length encodes time-varying data in the GPU. At each frame, we update view-aligned buckets that store the voxels contributing to each sheet-buffer. If the camera moves, the buckets are sorted in the GPU, otherwise they are directly used to compute the sheet-buffers. Our method supports interactive exploration of the data through time as well as camera and transfer function changes.