Visualization of big SPH simulations via compressed octree grids

Abstract

Abstract-Interactive and high-quality visualization of spatially continuous 3D fields represented by scattered distributions of billions of particles is challenging. One common approach is to resample the quantities carried by the particles to a regular grid and to render the grid via volume ray-casting. In large-scale applications such as astrophysics, however, the required grid resolution can easily exceed 10K samples per spatial dimension, letting resampling approaches appear unfeasible. In this paper we demonstrate that even in these extreme cases such approaches perform surprisingly well, both in terms of memory requirement and rendering performance. We resample the particle data to a multiresolution multiblock grid, where the resolution of the blocks is dictated by the particle distribution. From this structure we build an octree grid, and we then compress each block in the hierarchy at no visual loss using wavelet-based compression. Since decompression can be performed on the GPU, it can be integrated effectively into GPU-based out-of-core volume ray-casting. We compare our approach to the perspective grid approach which resamples at run-time into a view-aligned grid. We demonstrate considerably faster rendering times at high quality, at only a moderate memory increase compared to the raw particle set.

A framework for GPU‐accelerated exploration of massive time‐varying rectilinear scalar volumes

“…Output-sensitive approaches require adaptive loading of compressed data stored in out-of-core structures, which range, for static datasets, from a single-resolution set of compressed bricks [TBR * 12] to multiresolution structures such as octrees [CNLE09,Eng11,GIM12,RTW13] or hierarchical grids of bricks [HBJP12,FSK13]. In this context, a scalable preprocessing method, compressed streaming, and fast on-demand spatially independent decompression on the GPU, are required for maximum benefits [FM07].…”

State‐of‐the‐Art in GPU‐Based Large‐Scale Volume Visualization

“…Despite the hierarchical nature of these data structures, many early approaches assume that the entire volume fits into memory [LHJ99, WWH*00, BNS01]. Modern GPU approaches support traversing octrees directly on the GPU [GMG08, CNLE09, CN09, GGM10, RTW13], which is usually accomplished via standard traversal algorithms adapted from the ray‐tracing literature [AW87, FS05, HSHH07, PGS*07, HL09].…”

“…Tree traversal is based on an adapted kd-restart algorithm [FS05]. Engel [Eng11] and Reichl et al [RTW13] use the same basic structure and traversal method.…”

“…It also uses the kd-restart algorithm for octree traversal. Reichl et al [RTW13] employ a similar ray-guided approach, but target large smooth particle hydrodynamics (SPH) simulations.…”

Sparse PDF Volumes for Consistent Multi-Resolution Volume Rendering

“…The most common approach for handling large data in volume visualization is to use multi-resolution techniques [38], usually utilizing hierarchical data structures such as octrees [7, 14, 20, 31, 37] or 3D mipmaps [11, 16, 21]. These representations store iteratively pre-filtered and down-sampled versions of the original volume at discrete resolution levels.…”