ScaleFree: Dynamic KDE for Multiscale Point Cloud Exploration in VR
Lixiang Zhao 1 \href https://orcid.org/0000-0001-6181-1673, Fuqi Xie 1 \href https://orcid.org/0009-0008-4728-9346, Tobias Isenberg 2 \href https://orcid.org/0000-0001-7953-8644, Hai-Ning Liang 3 \href https://orcid.org/0000-0003-3600-8955, Lingyun Yu 1 \href https://orcid.org/0000-0002-3152-2587 · 2026
Read ScaleFree: Dynamic KDE for Multiscale Point Cloud Exploration in VR by Lixiang Zhao 1 \href https://orcid.org/0000-0001-6181-1673, Fuqi Xie 1 \href https://orcid.org/0009-0008-4728-9346, Tobias Isenberg 2 \href https://orcid.org/0000-0001-7953-8644, Hai-Ning Liang 3 \href https://orcid.org/0000-0003-3600-8955, Lingyun Yu 1 \href https://orcid.org/0000-0002-3152-2587 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-20758.txt.
Introduction
Lixiang Zhao and four co-authors build a tool for exploring huge point clouds in virtual reality. Their 2026 paper presents ScaleFree, a GPU-accelerated method based on kernel density estimation. A point cloud is a set of points in space, such as particles in a simulation. Kernel density estimation turns those points into a smooth density field. The method adapts as the user zooms across scales.
The target users work with massive scientific data. Cosmological simulations, for example, can hold billions of particles. Their structures span many scales at once, from voids to dense clumps. ScaleFree aims to keep exploration interactive at that size. It lets researchers move through data in VR and see structure at each scale. That speed turns a static dataset into a place one can inspect up close.
Catalog record
| CREATOR | Lixiang Zhao 1 \href https://orcid.org/0000-0001-6181-1673, Fuqi Xie 1 \href https://orcid.org/0009-0008-4728-9346, Tobias Isenberg 2 \href https://orcid.org/0000-0001-7953-8644, Hai-Ning Liang 3 \href https://orcid.org/0000-0003-3600-8955, Lingyun Yu 1 \href https://orcid.org/0000-0002-3152-2587 |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-20758
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
ScaleFree: Dynamic KDE for Multiscale Point Cloud Exploration in VR Lixiang Zhao 1 \href https://orcid.org/0000-0001-6181-1673, Fuqi Xie 1 \href https://orcid.org/0009-0008-4728-9346, Tobias Isenberg 2 \href https://orcid.org/0000-0001-7953-8644, Hai-Ning Liang 3 \href https://orcid.org/0000-0003-3600-8955, Lingyun Yu 1 \href https://orcid.org/0000-0002-3152-2587 Abstract We present ScaleFree, a GPU-accelerated adaptive Kernel Density Estimation (KDE) algorithm for scalable, interactive multiscale point cloud exploration. With this technique, we cater to the massive datasets and complex multiscale structures in advanced scientific computing, such as cosmological simulations with billions of particles. Effective exploration of such data requires a full 3D understanding of spatial structures, a capability for which immersive environments such as VR are particularly well suited. However, simultaneously supporting global multiscale context and fine-grained local detail remains a significant challenge. A key difficulty lies in dynamically generating continuous density fields from point clouds to facilitate the seamless scale transitions: while KDE is widely used, precomputed fields restrict the accuracy of interaction and omit fine-scale structures, while dynamic computation is often too costly for real-time VR interaction. We address this challenge by leveraging GPU acceleration with k-d-tree-based spatial queries and parallel reduction within a thread group for on-the-fly density estimation. With this approach, we can recalculate scalar fields dynamically as users shift their focus across scales. We demonstrate the benefits of adaptive density estimation through two data exploration tasks: adaptive selection and progressive navigation. Through performance experiments, we demonstrate that ScaleFree with GPU-parallel implementation achieves orders-of-magnitude speedups over sequential and multi-core CPU baselines. In a controlled experiment, we further confirm that our adaptive selection technique improves accuracy and efficiency in multiscale selection tasks. \onlineid 1680 \vgtccategory Research \vgtcinsertpkg \teaser ScaleFree dynamically recomputes density fields on the GPU, enabling on-the-fly density estimation of regions of interest (ROI) at any scale and position within massive multiscale point clouds in VR. keywords Multiscale visualization, kernel density estimation, interactive data selection, interactive data navigation, VR/AR/MR. 1 Introduction Advances in scientific computing have produced datasets that are not only larger in size but also richer in multiscale structural complexity. For example, modern cosmological simulations produce billions of particles, with details that span several orders of magnitude, to capture the formation and evolution of cosmic structures [ 55 , 54 ] . The fact that meaningful features often emerge at different scales requires analysts to fluidly transition between global and local perspectives—for instance, from cosmic filaments and clusters to fine-grained substructures in astronomical point clouds [ 55 , 28 ] . To understand the dynamics of structure formation, scientists need to visualize and explore the three-dimensional spatial patterns across multiple scales. Immersive environments delivered through virtual, augmented, and mixed reality head-mounted displays (HMDs) have emerged as superior tools for interpreting complex spatial relationships. Prior research has shown that immersive visualizations can enhance users’ understanding of complex spatial structures, including volume data [ 38 ] , astronomical data [ 74 ] , and node-link diagrams [ 27 , 60 ] . These benefits are particularly evident in navigation [ 53 ] and 3D manipulation tasks [ 8 ] , where frequent viewpoint changes are required to inspect spatial structures. The immersive environment not only provide stereoscopic perception of 3D structures but also support seamless transitions between global overviews and localized detail exploration. For instance, a user may zoom in to inspect a galaxy cluster and then zoom out to contextualize its dynamics within the surrounding large-scale structures. This process requires not only an immersive environment, such as VR, but also high-performance scientific computing and visualization systems capable of supporting free navigation across space, time, and scale. Achieving smooth transitions and effective exploration across multiple scales remains challenging, however, in particular when data must be transformed into an alternative representation to support further analysis. For example, KDE [ 19 ] has been widely used in point cloud visualization to uncover clusters and structural patterns in large multiscale datasets by converting discrete points into a continuous scalar density field. Such density fields are fundamental to spatial interaction, as they provide a robust basis for feature detection [ 15 , 45 , 50 , 12 ] and for constructing selection volumes [ 69 , 68 , 75 ] , thereby enabling users to identify, isolate, and manipulate regions of interest. Nonetheless, computing scalar density fields on the fly is computationally expensive and often fails to guarantee real-time interaction. Prior density-based selection techniques for point cloud exploration [ 68 , 69 , 75 ] have typically focused on selection interaction itself and relied on precomputed, single-scale density fields, rather than targeting dynamic multiscale recomputation during interaction. Such precomputation, however, implies that the density fields are static during exploration, constraining selection accuracy to the resolution defined a priori and leaving finer-scale structures unrepresented. This limitation becomes particularly pronounced in VR. Unlike desktop systems, VR presents users with a single, continuous egocentric view of the data space, where navigation, selection, and scale transitions are tightly coupled with head and body movement. Users frequently shift their focus between global overviews and localized regions, creating a strong demand for density fields that adapt immediately to scale and perspective changes. At the same time, VR imposes strict real-time constraints due to high-frequency stereoscopic rendering and low-latency interaction requirements. These challenges motivate us to explore density estimation techniques that can be dynamically computed at varying scales during real-time immersive exploration. We thus introduce ScaleFree, a fast, adaptive, and scalable KDE algorithm specifically optimized for interactive point cloud data analysis. Our method leverages GPU acceleration to perform on-the-fly density estimation, enabling scalar fields to be recomputed dynamically as users shift their focus across scales. By maintaining density fields that adapt repeatedly rather than relying on precomputed representations, ScaleFree facilitates more precise and responsive support for core exploration tasks such as spatial selection , feature localization , and multiscale navigation in large point cloud datasets. To evaluate the role of adaptive KDE in interactive VR exploration, we adopt a multi-layered evaluation strategy. We first compare our GPU-based implementation against CPU baselines to establish real-time feasibility, and then evaluate how integrating adaptive KDE affects selection effectiveness through technique demonstrations and a controlled user study, measuring accuracy, efficiency, and workload against state-of-the-art VR selection techniques. Specifically, we conducted a user study with 24 participants comparing three selection techniques that employ different density estimation approaches, including precomputed single-resolution density fields (PS), precomputed multi-resolution mipmap density fields (PM), and our dynamic ScaleFree approach. Results show that selections made with ScaleFree were faster and more accurate than those using precomputed density fields. Participants also reported lower workload and expressed a stronger preference for ScaleFree when conducting multiscale exploration tasks in VR. In summary, our GPU-accelerated adaptive KDE makes the following main contributions to large-scale spatial data exploration in immersive environments: • Adaptive and smooth scale transitions: enabling continuous transitions across scales without disruptive visual artefacts or noticeable delay (demonstrated in both selection \autoref subsec:selection:method and navigation \autoref subsec:navigation:method scenarios). • Stable viewpoint-driven exploration: supporting smooth, orientation-preserving transitions between overview and localized perspectives across scales during exploration of complex datasets (demonstrated in the navigation scenario, \autoref subsec:navigation:method). • Fast and responsive computation: improving selection accuracy and efficiency under varying scales, thereby supporting precise spatial interaction in immersive analytics (shown in the user study results, \autoref sec:userstudy). 2