学术文献库

Scientific visualization tools are currently not optimized to create cinematic, production-quality representations of numerical data for the purpose of science communication. In our pipeline \texttt{Estra}, we outline a step-by-step process from a raw simulation into a finished render as a way to teach non-experts in the field of visualization how to achieve production-quality outputs on their own. We demonstrate feasibility of using the visual effects software Houdini for cinematic astrophysical data visualization, informed by machine learning clustering algorithms. To demonstrate the capabilities of this pipeline, we used a post-impact, thermally-equilibrated Moon-forming synestia from \cite{Lock18}. Our approach aims to identify "physically interpretable" clusters, where clusters identified in an appropriate phase space (e.g. here we use a temperature-entropy phase-space) correspond to physically meaningful structures within the simulation data. Clustering results can then be used to highlight these structures by informing the color-mapping process in a simplified Houdini software shading network, where dissimilar phase-space clusters are mapped to different color values for easier visual identification. Cluster information can also be used in 3D position space, via Houdini's Scene View, to aid in physical cluster finding, simulation prototyping, and data exploration. Our clustering-based renders are compared to those created by the Advanced Visualization Lab (AVL) team for the full dome show "Imagine the Moon" as proof of concept. With \texttt{Estra}, scientists have a tool to create their own production-quality, data-driven visualizations.