🤖 AI Summary
This work addresses the poor dimensionality interpretability and low academic acceptance of static 3D scatter plots in scientific visualization—stemming from inconsistent projection methods and weak depth perception. We conduct a human-centered empirical study to optimize knowledge communication. Our method introduces design principles for static 3D scatter plots, implements a JavaFX framework supporting multi-channel visual encoding (color, size, opacity, and texture), and incorporates adaptive projection transformations to strengthen depth cues. We propose the first quantitative evaluation methodology combining multiple perceptual features to systematically assess how visual encodings affect reading time and dimensional interpretation accuracy. Experimental results show that our optimized plots reduce average reading time by 37% and improve 3D structural identification accuracy by 52%, significantly outperforming default 3D outputs from Matplotlib and MATLAB. The source code and evaluation protocol are publicly released.
📝 Abstract
Computationally and data intensive workloads including design space exploration or large studies often lead to multi-dimensional results, which are often not trivial to digest with conventional plotting software. 3D scatterplots can be a powerful technique to visualise and explore such datasets, especially with the help of colour mapping and other approaches to represent more than the 3 dimensions of the Cartesian coordinate system. However, modern software commonly lacks this multi-dimensional functionality or is ineffective. One limitation is the frequent use of isometric axes, which is equivalent to removing one entire dimension. In manuscripts, additional visual cues such as movement are also not present to mitigate for the loss of depth perception and spatial information, hence their relatively limited use as static figures. In this work, we present a novel open-source JavaFX-based plotting framework that focuses on easy exploration of multi-dimensional datasets, and provides unique features or feature combinations to improve knowledge transfer from single stand-alone plots. An empirical study was conducted within an academic institution to quantify the effectiveness of feature or feature combinations on 3D scatterplots in terms of reading time and accuracy.