Smooth Mixed-Resolution GPU Volume Rendering

Abstract

We propose a mixed-resolution volume ray-casting approach that enables more flexibility in the choice of downsampling positions and filter kernels, allows freely mixing volume bricks of different resolutions during rendering, and does not require modifying the original sample values. A C0-continuous function is obtained everywhere with hardware-native filtering at full speed by simply warping texture coordinates of samples in transition regions. Additionally, we propose a simple but powerful, flat texture packing scheme that supports mixing different resolution levels in a single 3D volume cache texture with a very simple and fast address translation scheme. Although this packing constrains full scalability, it enables mixing different resolution levels in GPU-based ray-casting with only a single rendering pass. We demonstrate our approach on large real-world data, obtaining a continuous scalar function and shading at brick boundaries, using single-pass ray-casting at real-time frame rates.