Abstract:
Understanding fluid dynamics is critical to many fields of scientific research ranging from aerodynamics to oceanography. To meet this  need, flow visualization has been a central topic of scientific visualization over the past three decades. Integral flow lines such as streamlines intuitively demonstrate the flow patterns, and therefore, become the most widely used visual means to represent and visualize flow fields. Most of the existing works focused on how to produce informative streamlines from flow fields via seed placement or streamline selection. However, the inverse problem of reconstructing flow fields as accurate as possible from streamlines is also intriguing but much less explored. In this talk, I will present a new approach for streamline-based flow field reconstruction. Our method can work in the in situ visualization setting by tracing streamlines from each time step of the simulation and storing compressed streamlines for post hoc analysis where users can afford longer reconstruction time for higher reconstruction quality using decompressed streamlines. At the heart of our approach is a deep learning method for vector field reconstruction that takes the streamlines traced from the original vector fields as input and applies a two-stage process to reconstruct high-quality vector fields.

Bio:
Jun Han is a PhD student at University of Notre Dame. He received a BS degree in software engineering and a MS degree in computer software and theory in 2014 and 2017, respectively. Both degrees are from Xidian University. His current research focuses on applying deep learning techniques to solve data visualization problems.