A Multi-ROI Camera Motion Exploration Approach for Enhancing Image-based Smart In-Situ Visualization

Image-based in-situ visualization has emerged as a flexible and powerful approach for batch-based, large-scale simulation and visualization by combining in-situ image generation on the HPC side with offline exploratory visual analysis on the local PC side. Several information entropy-based smart in-situ visualization methods have also been proposed so far to facilitate the analysis of large amounts of rendering images from different viewpoints or camera positions by extracting a set of potentially meaningful images that may aid in understanding the underlying simulated phenomena. However, existing methods have been limited to a single camera focus point or region of interest (ROI), which may not be sufficient to fully comprehend the complex, time-varying simulated phenomena. As an initial attempt to minimize this limitation, we investigate the use of multiple ROIs for enabling the visual tracking of other potentially meaningful regions in the simulated phenomena. For this purpose, we adopted the existing automatic focus point (ROI) and zoom level adjustment technique by enabling the search for additional ROIs and their corresponding zoom levels. Additionally, we employed a static initial viewpoint (camera position) setup for this initial attempt. Considering that the number of possible camera movements increases quadratically with the number of ROIs, we propose an offline post-processing method that enables interactive camera motion exploration based on user-defined parameters. We evaluated this approach with domain experts by using different CFD simulation codes on both a traditional x86 server and the Arm-based supercomputer Fugaku, and obtained promising preliminary results, which make us optimistic about further enhancements.