Winding angle assisted particle tracing in distribution-based vector field

Abstract

Distribution models are widely used for data reduction applications. e Gaussian mixture model (GMM) is a powerful tool to capture multiple-peak distributions. For distribution-based vector eld datasets represented by GMM, there are still loss of information which sometimes causes too much error when performing ow line tracing tasks. As a compensation, we analyze the vector transition pa ern between consecutive vector directions. e vector transition is depicted by distributions of winding angles. When performing streamline and pathline tracing, we utilize the winding angle to estimate a conditional distribution of local vectors, using the Bayes eorem. e conditional distribution can be used for both Monte Carlo ow line tracing, and single ow line tracing. We applied our distribution model on data reduction applications, and demonstrated that improved ow line tracing quality was achieved. CCS CONCEPTS •Computing methodologies →Parametric curve and surface models;

Efficient and Portable Distribution Modeling for Large-Scale Scientific Data Processing with Data-Parallel Primitives

“…For example, Dutta et al [1] and Thompson et al [3] divided a volume dataset into sub-blocks, where data values of each sub-block are stored by a univariant GMM or histogram when the spatial resolution of the dataset is huge. Li et al [17] used the same idea to model vectors in a sub-block with a multivariant GMM to store the vector dataset compactly. Figure 2a illustrates the above ideas.…”

“…It can represent a complicated distribution using a few parameters to provide a compact and accurate distribution representation. Therefore, GMM has been widely used to facilitate scientific data reduction and visualization [2,8,[17][18][19]37,38].…”

“…Our data-parallel primitive-based algorithms can facilitate the data process that is run on supercomputers with different types of computing nodes. In this work, we propose parallel multi-set distribution modeling algorithms for multi-variant histogram [3,[12][13][14][15] and GMM [1,2,4,[15][16][17] modeling, because these are the most popular non-parametric and parametric distribution representations in the scientific data modeling, respectively.…”