Contour forests: Fast multi-threaded augmented contour trees

Abstract

This paper presents a new algorithm for the fast, shared memory multi-threaded computation of contour trees on tetrahedral meshes. In contrast to previous multi-threaded algorithms, our technique computes the augmented contour tree. Such an augmentation is required to enable the full extent of contour tree based applications, including for instance data segmentation. Our approach relies on a range-driven domain partitioning. We show how to exploit such a partitioning to rapidly compute contour forests. We also show how such forests can be efficiently turned into the output contour tree. We report performance numbers that compare our approach to a reference sequential implementation for the computation of augmented contour trees. These experiments demonstrate the run-time efficiency of our approach. We demonstrate the utility of our approach with several data segmentation tasks. We also provide a lightweight VTK-based C++ implementation of our approach for reproduction purposes.

The Effect of Data Transformations on Scalar Field Topological Analysis of High-Order FEM Solutions

“…Finally, our analysis pipeline involves computing an overview of features with TTK, in particular extracting critical points and computing their persistence, as well as visualizing persistence diagrams and persistence curves. We use this information to guide a segmentation of the data domain based on using the contour forests method for fast extraction of contour trees [30]. We consider multiple scales of features, based on a manual analysis of the persistence diagram to set persistence thresholds.…”

“…Reeb graphs and contour trees have been well studied by the visualization community, offering a number of approaches for their computation in two-and three-dimensional settings [5,17,22,23,30,54,56,57,67]. In our work, we construct segmentations of the input domain based on the contour tree, and rely on the fact that these segmentations correspond to regions of interest in the data (specifically, we consider vortex identification).…”

Task-Based Augmented Contour Trees with Fibonacci Heaps

“…Among the approaches which addressed the time performance improvement of contour tree computation through shared-memory parallelism, only a few of them rely directly on the original merge tree computation algorithm [10], [11]. This algorithm is then used within partitions of the mesh resulting from a static decomposition on the CPU cores, by either dividing the geometrical domain [55] or the scalar range [52]. This leads in both cases to extra computation (with respect to the sequential mono-partition computation) at the partition boundaries when joining results from different partitions.…”

“…This leads in both cases to extra computation (with respect to the sequential mono-partition computation) at the partition boundaries when joining results from different partitions. This can also lead to load imbalance among the different partitions [52].…”

A generic framework for the structured abstraction of images

“…In this paper, we refer to this tree structure as the topographic map and we use the Fast Level Set Transform (FLST) [Monasse and Guichard, 2000] to compute it e ciently. Alternatively, one might choose contour tree computation methods, widely used in the scienti c visualization community [Carr et al, 2003, Gueunet et al, 2016. A schematic representation of the topographic map is displayed in Figure 2.…”