Publications
Vortex Boundary Identification using Convolutional Neural Network
IEEE Visualization Short Papers, 2020
Abstract
Feature extraction is an integral component of scientific visualization, and specifically in situations in which features are difficult to formalize, deep learning has great potential to aid in data analysis. In this paper, we develop a deep neural network that is capable of finding vortex boundaries. For training data generation, we employ a parametric flow model that generates thousands of vector field patches with known ground truth. Compared to previous methods, our approach does not require the manual setting of a threshold in order to generate the training data or to extract the vortices. After supervised learning, we apply the method to numerical fluid flow simulations, demonstrating its applicability in practice. Our results show that the vortices extracted using the proposed method can capture more accurate behavior of the vortices in the flow.
Download
preprint
video
doiVideo
BibTeX
@inproceedings{Berenjkoub20VisShort,
author = {Berenjkoub, Marzieh and Chen, Guoning and G{\"u}nther, Tobias},
title = {Vortex Boundary Identification using Convolutional Neural Network},
booktitle = {IEEE Visualization Short Papers},
pages = {261-265},
year = {2020},
publisher = {IEEE},
address = {Utah, US},
doi = {10.1109/VIS47514.2020.00059},
}
