Scalable Hybrid Unstructured and Structured Grid Raycasting

Abstract

This paper presents a scalable framework for real-time raycasting of large unstructured volumes that employs a hybrid bricking approach. It adaptively combines original unstructured bricks in important (focus) regions, with structured bricks that are resampled on demand in less important (context) regions. The basis of this focus+context approach is interactive specification of a scalar degree of interest (DOI) function. Thus, rendering always considers two volumes simultaneously: a scalar data volume, and the current DOI volume. The crucial problem of visibility sorting is solved by raycasting individual bricks and compositing in visibility order from front to back. In order to minimize visual errors at the grid boundary, it is always rendered accurately, even for resampled bricks. A variety of different rendering modes can be combined, including contour enhancement. A very important property of our approach is that it supports a variety of cell types natively, i.e., it is not constrained to tetrahedral grids, even when interpolation within cells is used. Moreover, our framework can handle multi-variate data, e.g., multiple scalar channels such as temperature or pressure, as well as time-dependent data. The combination of unstructured and structured bricks with different quality characteristics such as the type of interpolation or resampling resolution in conjunction with custom texture memory management yields a very scalable system.

Interactive Volume Visualization of General Polyhedral Grids

“…This is a significant limitation, since the cell sizes in unstructured grids used for simulation can vary by several orders of magnitude. Muigg et al [32] have introduced the first GPU ray-caster that renders more general cells directly. However, for reasons of memory addressing and alignment, their overall data layout targets grids with only a small number of different cell types.…”

“…Such bricking methods have also been applied in the context of unstructured meshes [32,33,40]. Progressive rendering and streaming is also used to cope with large grids [9].…”

“…Reasons are: scalability, memory management, adaptivity with respect to characteristics of the contained cells, coarse-grained culling, and the potential for direct parallelization across multiple GPUs. The brick geometry corresponds to the leaves of a kD tree, which is built using criteria similar to previous work [32]. However, we avoid explicitly clipping cells against brick boundaries.…”

“…This can be done very efficiently by encoding IDs in colors and rasterizing the grid boundary accordingly. In previous GPU ray-casting approaches [13,32,42,44], the triangle mesh representing the volume boundary had to be stored explicitly in a format suitable for fast rasterization (e.g., in OpenGL vertex buffer objects). We avoid permanently storing this redundant representation for the whole mesh.…”

“…Existing visualization systems that are not constrained to just tetrahedra still only support a small number of convex cell types [32], or essentially perform point splatting [48]. In this paper, we introduce the first interactive visualization approach for extremely complex unstructured grids, which takes into account all of the properties of stateof-the-art volume meshes outlined above.…”