Out-of-core volume rendering for time-varying fields using a space-partitioning time (SPT) tree

Abstract

In this paper, we propose a novel out-of-core volume rendering algorithm for large time-varying fields. Exploring temporal and spatial coherences has been an important direction for speeding up the rendering of time-varying data. Previously, there were techniques that hierarchically partition both the time and space domains into a data structure so as to re-use some results from the previous time step in multiresolution rendering; however, it has not been studied on which domain should be partitioned first to obtain a better re-use rate. We address this open question, and show both theoretically and experimentally that partitioning the time domain first is better. We call the resulting structure (a binary time tree as the primary structure and an octree as the secondary structure) the spacepartitioning time (SPT) tree. Typically, our SPT-tree rendering has a higher level of details, a higher re-use rate, and runs faster. In addition, we devise a novel cut-finding algorithm to facilitate efficient out-of-core volume rendering using our SPT tree, we develop a novel out-of-core preprocessing algorithm to build our SPT tree I/O-efficiently, and we propose modified error metrics with a theoretical guarantee of a monotonicity property that is desirable for the tree search. The experiments on datasets as large as 25GB using a PC with only 2GB of RAM demonstrated the efficacy of our new approach.

An Animated Visualization Method for Large-Scale Unstructured Unsteady Flow

“…Shen et al [7] designed a time-space partitioning (TSP) tree that divides the spatial domain into an octree as the primary structure, then divides the temporal domain into a binary time tree as the secondary structure and uses the data in the adjacent spatiotemporal domain to represent the current data based on the error tolerance. Du et al [8] proposed a space-partitioning time (spt) tree, which first divides the temporal domain into a fully balanced binary time tree as a primary structure and then divides the spatial domain into a standard complete octree, obtaining a higher data reuse rate because unsteady data exhibits a stronger correlation in time. Ma et al [9] used temporal consistency to prune tree nodes by storing each time step of data in an octree form separately and then comparing the similarity of adjacent time step data for pruning.…”

Graphs in Scientific Visualization: A Survey

“…Du et al . [DCS09] studied the open question of which domain should be partitioned first for a better data reuse rate. They showed both theoretically and experimentally that partitioning the time domain first is better.…”

State‐of‐the‐Art in Compressed GPU‐Based Direct Volume Rendering

“…In this framework, indexed key frames plus differential encoding of arbitrary time steps can theoretically be used to provide better access to random points in the sequence. In such approaches it is also common to exploit temporal coherence by first performing a spatial subdivision and, then, compactly encoding corresponding blocks of adjacent time steps [DCS09].…”