GRIVAPP 2026 Abstracts

Area 1 - Geometry — Modeling and Rendering
Full Papers Short Papers
Area 2 - Information Visualization & Visual Analytics
Full Papers Short Papers
Area 3 - Interactive Environments and Human-Computer Interaction
Full Papers Short Papers
Area 4 - Scientific Visualization
Full Papers Short Papers
Area 5 - Animation and Simulation
Full Papers Short Papers

Area 1 - Geometry — Modeling and Rendering

Full Papers
Paper Nr: 12
Title:

Direct Mesh Booleans: A Step towards Non-Restrictive Boolean Operations
Authors:

Roman Čižmarik and Michal Španěl
Abstract: In this paper, we propose a robust algorithm for Boolean operations on triangle meshes that imposes no input restrictions. Boolean operations are a fundamental tool in 3D modeling and geometry processing. Yet, existing approaches are prone to errors when faced with common defects in triangle meshes, such as open boundaries, non-manifold geometry, and self-intersections, or they disallow such inputs entirely. Exactness and numerical stability of our method are ensured by leveraging recent advances in exact computations. The method operates directly on the input meshes without any volumetric transformation, thus providing accurate conformity with the input surfaces. A lightweight regularization guarantees two-manifold outputs, albeit permitting open boundaries if necessary. Extensive evaluation on thousands of real-world meshes from the Thingi10k dataset demonstrates that our method’s reliability, robustness, and speed outperform prior art. The implementation is available as open-source from https://github.com/RomanCizmarik/Direct-Mesh-Booleans.git.
Download
Paper Nr: 24
Title:

X4Rate: A Scene-Adaptive Extrapolation Framework for 4× Frame Rate Upsampling
Authors:

Akanksha Dixit and Smruti R. Sarangi
Abstract: High-frequency displays are widely used in gaming and VR; sadly, GPUs often fail to sustain high frame rates, leading to visual artifacts like judder and motion blur. State-of-the-art approaches to enhance frame rates include interpolation and extrapolation. Interpolation typically yields higher-quality results because it leverages both past and future frames. However, it is more suitable for doubling frame rates, as further upsampling requires multiple passes, necessitating repeated forward and backward frame references, which introduces unwanted latency and complexity. Extrapolation, on the other hand, can continuously generate new frames at high rates. In this work, we aim to increase the frame rate by 4×, making extrapolation the preferred choice. However, no single extrapolation algorithm performs consistently well across all scene types. Each one is tailored to specific scenarios and exhibits strengths and weaknesses depending on the context. We propose X4Rate, a frame generation framework that combines multiple existing methods to deliver results superior to any individual algorithm by selecting the best algorithm for the current scene. It is basically a decision predictor (meta-learner) that dynamically selects the best-performing frame extrapolation algorithm at runtime. This adaptive selection ensures high visual quality consistently because when one method underperforms, others compensate. X4Rate achieves 4× the rendering rate while delivering high-quality frames, outperforming the nearest competitor, ExtraNet, with a 22.13% improvement in the PSNR.
Download
Paper Nr: 53
Title:

Evaluation of Novel View Synthesis in the Context of Radiometric Drone Imagery
Authors:

Christoph Praschl, Sebastian Schoibesberger and David C. Schedl
Abstract: Recent advancements in neural scene representations, specifically Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have revolutionized novel-view synthesis. However, these methods are predominantly optimized for and evaluated on ground-based, visible-spectrum (RGB) data, leaving a significant gap in their application to aerial, radiometric thermal imagery, which is critical for industrial inspection, search-and-rescue, and environmental monitoring. In this work, we systematically investigate the applicability of these state-of-the-art paradigms to radiometric thermal imagery acquired from airborne drone platforms. We introduce a novel, publicly available multimodal dataset captured using a DJI M30T system, comprising synchronized RGB and radiometric thermal frames of a building. We conduct a comprehensive evaluation comparing specialized thermal approaches (ThermalNeRF, ThermoNeRF, Thermal3DGS) against general-purpose methods (nerfacto, gsplat). Our assessment utilizes a suite of quantitative metrics (PSNR, SSIM, MAE, LPIPS, and DISTS) complemented by qualitative visual analysis. Results indicate that Thermal3DGS achieves state-of-the-art performance in the thermal domain (PSNR 22.99, SSIM 0.845), effectively mitigating artifacts common in low-texture thermal data. Conversely, gsplat demonstrates superior RGB synthesis and competitive thermal performance, suggesting that general-purpose splatting representations are robust enough for cross-spectral applications. This work bridges the gap between aerial radiometric sensing and neural rendering, demonstrating that off-the-shelf drone thermography can be utilized for high-fidelity 3D thermal reconstruction with minimal adaptation.
Download
Paper Nr: 60
Title:

Pose-Aware Multi-View Diffusion for Consistent 3D Texture Generation from a Single RGB Image
Authors:

Jierong Wen, Jit Chatterjee, Gijs Fiten and Maria Torres Vega
Abstract: Generating realistic and view-consistent textures for a 3D mesh from a single reference image is a challenging task, as the missing surface appearance must be inferred without prior knowledge of multi-view observations. While recent diffusion-based approaches have improved texture synthesis, they often suffer from multi-view inconsistency due to the lack of 3D awareness. In this work, we introduce a pose-aware multi-view diffusion model that augments a standard depth-to-image diffusion pipeline with camera pose labels and directional prompts as additional conditions, enabling the model to generate viewpoint-coherent rendered images. The model is fine-tuned on a multi-view rendered dataset and integrated into a coarse-to-fine texture generation pipeline, where synthesized views are progressively back-projected into UV space and completed via iterative inpainting. Experiments on the 3D-FUTURE dataset demonstrate that our approach improves cross-view semantic consistency and visual fidelity in both multi-view image generation and full texture reconstruction, achieving clear reductions in FID and KID over existing pipelines.
Download
Short Papers
Paper Nr: 16
Title:

Focal Effects Rendering Using Gaussian Mixture Models
Authors:

Julia Jodczyk and Łukasz Dąbała
Abstract: One of the challenges of light transport simulation is the so-called local effects, which are caused by focal points - small areas where many light paths intersect. Scenes with many focal effects are difficult to render because there is a low probability of finding them through random sampling. Instead, a more targeted sampling distribution is required. We propose a new path guiding method based on a Gaussian mixture model. The model learns the spatial distribution of light throughout the scene, which allows identification of focal points and guiding light rays towards them. The proposed method is evaluated using a set of test scenes containing different types of local effects and is compared to standard path tracing and selected path guiding algorithms. The results show that the algorithm is capable of finding clusters of high light intensity and achieves competitive results in terms of the quality of rendered scenes.
Download
Paper Nr: 26
Title:

Adaptive Meshes with Encoded Services
Authors:

Ben Lee and Stephen Brooks
Abstract: 3D meshes consume more storage than is required for most applications. In this paper we use this surplus of precision to connect to a variety of online services, all while still being able to use the mesh as-is in the graphics pipeline. We describe our method to link to online services through an encoded URL in the mesh itself. We then introduce our machine learning based subdivision surface approach which uses the extra precision to also store user driven preferences. We then describe a method to store an embedding that can be used to retrieve similar meshes from an online database. Our similarity model does not rely on model categories and our embeddings are also significantly smaller, since they are encoded directly into the mesh.
Download
Paper Nr: 46
Title:

RGB ReSTIR: Decorrelating Spatiotemporal Importance Resampling with Per-Channel Reservoirs
Authors:

Markku Mäkitalo, Saku Haikio, Julius Ikkala, Alessandro Foi and Pekka Jääskeläinen
Abstract: ReSTIR is a family of state-of-the-art spatiotemporal resampling algorithms utilized for improving the efficiency of photorealistic rendering. In particular, ReSTIR PT is commonly used for accelerating path tracing, a method that enables a high level of photorealism through a Monte Carlo based approximation of the global illumination. However, ReSTIR PT produces correlated samples due to its prominent reuse of spatially and temporally close pixels in the sample reservoirs, which typically manifests as visible color noise. ReSTIR-based algorithms typically only use luminance data for estimating their resampling target function, which means that in general, they cannot converge to a fully decorrelated image even with large reservoir sizes. In this paper, we present RGB ReSTIR, a multichannel variant of ReSTIR PT that maintains separate reservoirs and estimates separate target functions for each color channel. This approach allows the resampling to produce images with significantly less color noise than ReSTIR PT, especially for scenes with complex colored lighting. We demonstrate that RGB ReSTIR is able to converge towards a fully decorrelated image as the maximum confidence is increased (i.e., with longer temporal reservoir history), typically reaching an order of magnitude lower average sample autocovariance than ReSTIR PT.
Download
Paper Nr: 48
Title:

Context-Aware DeepSVG: A Model to Address Loss of Detail and Meaningless Geometries in Generative Vector Graphics
Authors:

Edwin Zheng Yuanyi
Abstract: This project explores techniques to extend the DeepSVG model, used to recreate scalable vector graphics (SVGs) and generate animations, to address open problems in the field of vector graphics deep learning. The first is the Loss of Detail problem, where small features vanish in recreations. The second is the Meaningless Geometries problem, where vector graphics are generated with shapes that do not contribute to the overall appearance. To measure the effectiveness of each proposal, this paper defines metrics that evaluate the sensibility of recreated SVGs and how effectively each model addresses the research problems. The first approach involved computing a regularization term based on how well the model addresses both problems. The second approach involved tackling a related problem, minimizing the presence of overlapping shapes in the outputs. New models were produced from these approaches and compared with the base model.
Download
Paper Nr: 70
Title:

Optimizing Neural Path Guiding with Parametric Mixture Models: A Comprehensive Evaluation and Refinement
Authors:

Paolo Russo and Lorenzo Mauro
Abstract: This paper offers a comprehensive analysis of Neural Path Guiding with Parametric Mixture Models, refining and building upon state-of-the-art techniques to address the challenges of light path sampling in Physically Based Rendering. While Monte Carlo Integration is a cornerstone technique for rendering, it often faces difficulties in efficiently sampling light paths in complex lighting scenarios, such as those involving indirect lighting and caustics. Our work critically examines and optimizes existing methodologies, mainly through an improved configuration of Neural Path Guiding that achieves state-of-the-art performance. We employ Hash Grid and Spherical Harmonics encodings for input embeddings alongside a corrected implementation of the Normalized Anisotropic Spherical Gaussian distribution, enhanced by a novel activation strategy. We further identify the dimension of the training batch size as a critical hyperparameter for the method’s success. An extensive ablation study explores the impact of various hyperparameters, offering more profound insights into the neural network’s behavior. Our results show notable improvements in noise reduction across various scenarios, validating the effectiveness of our approach and its potential to push the boundaries of realistic and efficient rendering.
Download
Paper Nr: 25
Title:

Encoding Functionality Directly into Polygonal Models
Authors:

Ben Lee and Stephen Brooks
Abstract: Vertices in 3D meshes are typically stored with more precision than is needed for most use cases. We argue that this unused precision could be used to enhance the functionality of meshes by encoding additional metadata in the mesh itself. And with this approach, the user of the mesh can choose to utilize the embedded functionality without requiring any extra preprocessing steps, while those who ignore it can use the mesh normally. We apply this approach to two different use cases for the purpose of augmenting graphics functionality. We first show that we can store ambient occlusion to a level which is imperceptible to its vertex colored equivalent. We then use directly encoded information into the mesh to create a prefractured version of the mesh when subjected to impacts.
Download
Paper Nr: 74
Title:

A Survey of Roughness-Based Applications Intended for Characterising, Controlling or Generating Real or Synthetic Non-Smooth Surfaces
Authors:

Mohamad Janbein, Céline Roudet, Christian Gentil, Clément Poull and Lucie Druoton
Abstract: Surface roughness is an inherently complex notion for which no unique definition exists, as its meaning de-pends on the considered application domain. This difficulty has led different fields to develop their own interpretations, characterisations, generations (and even tools) to address roughness. This is an essential concept in Computer Graphics (CG), when it comes to modelling 3D content for realistic scene creation, subject to an evaluation of their visual quality after undergoing geometric operations (e.g., simplification, compression, remeshing, or smoothing). It is also central in other fields, such as industry or engineering, where its characterisation and control are essential for design, manufacturing, and quality control. This survey examines the most significant works on roughness, exploring the various ways it is defined, utilized, modeled, and generated across the considered fields. It highlights how these initiatives differ and complement each other, with the hope of inspiring the development of new trends in CG or geometric modelling.
Download

Area 2 - Information Visualization & Visual Analytics

Full Papers
Paper Nr: 21
Title:

Dynamic Tree Colors: Adaptive Discriminable Hierarchies with Minimum Instability
Authors:

Tobias Mertz, Steven Lamarr Reynolds and Jörn Kohlhammer
Abstract: Hierarchical color maps can support users in the analysis of hierarchical data. For large hierarchies, dynamic color maps can improve discriminability upon user interactions, but the incremental color changes may cause users to lose their orientation in the data set. To address this challenge, we present Dynamic Tree Colors, a dynamic hierarchical color map that can be configured to a suitable tradeoff between discriminability and color stability. We also define quality metrics for both criteria and investigate our algorithm’s performance with respect to these metrics as well as a user study with 18 participants. Our results indicate that Dynamic Tree Colors yields good results in a wide range of application scenarios, but it does not achieve the performance of the state-of-the-art algorithm Cuttlefish in the specific scenario that algorithm was designed for.
Download
Paper Nr: 34
Title:

Model-Aware Visual Analytics for Aligning Data Shift in Network Traffic Classification
Authors:

Igor Cherepanov, David Sessler, Alexander Feil, Alex Ulmer and Jörn Kohlhammer
Abstract: Deep learning (DL) models that excel on training data often underperform in deployment due to distribution shift, which is a change in the properties of the input data between the training environment and real-world conditions. Existing tools provide coarse drift scores but offer limited support for localization, interpretation, and remediation. Reliable classification is essential in the critical domain of network traffic, which is important for cybersecurity and network management. We present a human-centered visual analytics system for network traffic classification that integrates a DL classifier, post-hoc out-of-distribution (OOD) detection methods, and interactive analysis. Our approach systematically benchmarks OOD detectors, integrates the most effective one to a strong DL baseline, and embeds it into a dashboard that enables experts to explore drift phenomena, from global distributional changes down to individual packets. The visual-interactive system combines model-aware projections, score distributions, class-level summaries, and local explainable artificial intelligence (XAI) to support drift localization and model interpretation. A qualitative evaluation shows how the system helps practitioners detect shifts, contextualize them with domain expertise, and plan corrective actions. By bridging automated detection with interactive exploration, our work advances more trustworthy and adaptive AI systems for dynamic network environments.
Download
Paper Nr: 44
Title:

APPA: A Cluster-Preserving Approximating Parametric Projection Algorithm
Authors:

Alister Machado and Alexandru Telea
Abstract: Dimensionality Reduction (also called projection) is the tool of choice for visualizing high-dimensional data due to its applicability to datasets of different kinds, sizes, and dimensionalities. However, many projection algorithms scale poorly with dataset size, limiting their applicability on large datasets. A particularly successful approach to scalability is to approximate the projection by computing it for a subset of the data, and use the (fast) approximation to project the full dataset. Neural Network Projection (NNP) is one such approximation algorithm which is fast at both training and inference, is projection- and dataset-agnostic, has out-of-sample ability, and is simple to implement. Yet, NNP creates projections in which data points diffuse over the projection space even if they were clustered at training time. Since groupings are crucial features of projections this severely limits NNP’s attractivity as an alternative projection technique. We propose APPA (Approximating Parametric Projection Algorithm), a refinement of NNP that inherits all of NNP’s qualities while strongly reducing the diffusion problem. We evaluate APPA across a variety of datasets and projection techniques, demonstrating its ability to maintain the quality of the reference projection.
Download
Paper Nr: 61
Title:

Graph Visualization Design Guidelines as Learnable Predicates
Authors:

Sjoerd Vink, Brian Montambault, Mingwei Li, Remco Chang and Michael Behrisch
Abstract: Graphs are widely used to represent complex, interconnected data across domains, yet choosing an effective visualization remains difficult because existing design knowledge is fragmented and inconsistent. This lack of a unified foundation prevents researchers from integrating findings into a cumulative body of knowledge, leaving valuable results isolated. It also hinders designers and practitioners, who cannot readily translate such findings into actionable strategies for their own goals and contexts. We propose a predicate-based representation that formalizes visualization guidelines as bounded conditions over descriptive graph statistics. Predicates directly mirror the qualitative structure of design guidelines. For example, a rule might specify that if graph density is low, a node-link diagram is appropriate, whereas if density is high, an adjacency matrix should be used. Unlike static handcrafted rules, they can also be learned, optimized, and adapted as new findings or usage contexts emerge. As a result, fragmented knowledge is consolidated into a formal and extensible foundation for graph visualization design and recommendation. We evaluate this approach by testing its ability to (i) recover expert rules and (ii) adapt to user-specific preferences while generalizing to unseen graphs. The results show that the learned predicates closely reproduce expert-derived guidelines, accommodate diverse preference patterns, and achieve strong performance on held-out data, demonstrating a promising path toward more systematic and cumulative graph visualization design knowledge.
Download
Short Papers
Paper Nr: 14
Title:

Towards Interpretable Multimodal Embeddings: A QR-Based Prototype Projection Approach
Authors:

João Vitor T. Leal, Artur A. Oliveira, Mateus Espadoto, Roberto M. Cesar Jr. and Roberto Hirata Jr.
Abstract: Multimodal embedding models trained on paired image-text data have enabled powerful zero-shot classification and cross-modal retrieval by learning a shared representation space. However, in high-dimensional embedding spaces, cosine similarity often fails to reflect true semantic relationships, producing unreliable similarity scores across semantically distinct instances. In this work, we propose a simple and model-agnostic method to improve the structure and interpretability of such embedding spaces. Given any text-image aligned embedder, we define class prototype embeddings based on textual descriptions, and compute a QR decomposition of these prototypes to obtain an orthonormal basis aligned with the class semantics. Image embeddings are then projected into this subspace, producing representations with improved alignment to class prototypes. Our experiments demonstrate that this transformation enhances clustering quality and embedding structure, with classification performance remaining effectively unchanged, and requires no retraining of the original model. The method is applicable to any pretrained multimodal embedder, providing a simple geometric enhancement that supports better structured downstream tasks such as classification, clustering, and outlier detection.
Download
Paper Nr: 28
Title:

MeDiTwin: A Platform for a Secure Medical Digital Twin for Data Sharing and Analysis
Authors:

Salmah Ahmad, Bianca Bartelt, Matthias Enzmann, Jörn Kohlhammer, Stefan Wesarg and Ruben Wolf
Abstract: This paper presents MeDiTwin, a secure medical digital twin (MDT) platform designed to enhance patient data management within healthcare systems while ensuring robust data protection. As medical digital twins gain traction, their potential for enabling personalized treatment planning and advanced clinical research becomes increasingly evident. However, the inherent sensitivity of medical data requires strict adherence to EU General Data Protection Regulation (GDPR) standards, compelling the development of a platform that offers granular, patient-centric access control. MeDiTwin utilizes advanced cryptographic techniques, namely attribute-based encryption (ABE), to ensure data confidentiality and integrity during exchanges across institutional boundaries. The platform’s usability and feasibility are validated through expert feedback and structured user studies, which highlight its efficacy in managing complex medical data scenarios. This paper outlines the system architecture, usability metrics, and insights from expert interviews, emphasizing the transformative potential of secure digital twin technologies in enhancing healthcare delivery while safeguarding patient autonomy and data security.
Download
Paper Nr: 49
Title:

Building a 3D Mini-Map from Reality Captures and Agentic AI
Authors:

Ben Falchuk
Abstract: The rapid rise of Artificial Intelligence (AI) tools in recent years has resulted in exciting new possibilities for computer programmers, game and web developers, and three-dimensional (3D) art practitioners. In particular, algorithms that use AI techniques to infer structure from a series of still images, have very quickly had a profound effect upon the landscape of computer and web graphics, and can now be found in domains including cinematic visual effects, architecture, design, advertising, and heritage. Of interest in this study are what we call 3D mini-maps (3DMM): stylized reality captures that enable interactive, full-perspective exploration. This position paper takes the following two stances: 1) creating 3DMM’s from scratch is both tedious and difficult, requiring technical and design skills, but still offers high rewards; and 2) Large Language Model agents are rapidly gaining traction as a powerful means of automation and could be turned loose on the task of generating a 3DMM.
Download
Paper Nr: 59
Title:

Eye-Tracking Support for Adaptive Document Exploration: Design Space Model and Application Examples
Authors:

M. Tytarenko, C. Söls, D. Atzberger, S. Lengauer and T. Schreck
Abstract: Adaptive visualization systems often rely on user interaction data to personalize content presentation as well as improve user experience. We introduce a gaze-driven design space linking interface components (e.g., full text, word clouds, infographics), gaze metrics (fixations, dwell time, level of detail), and usage scenarios (history reconstruction, attention summarization, recommendation). We present two gaze-based visualizations as a part of our Gaze-Adaptive Dashboard as both data capture tools and as reflective interfaces: (1) a fixation timeline for the reconstruction of exploration histories, (2) a hierarchical bar chart illustrating attention across interface segments, and furthermore, we provide a recommender system, presenting adaptive recommendations based on gaze-based preferences. While our example is based on a Consumer Health Information System, the proposed approach is domain-independent and can aid any adaptive web application that benefits from visual analytics of user attention. We lay out the design trade-offs that emerge in visualizing noisy gaze data, handling incidental versus meaningful fixations, and integrating summarization approaches through Large Language Models (LLMs). Our paper highlights the potential of attention-aware visualizations to support more fine-grained user modeling and adaptive interface design, and reveals open challenges for future work at the intersection of interaction, visual analytics, and personalization.
Download
Paper Nr: 69
Title:

Growlytics: A Data-Driven, Rule-Based Planner for Personalized Garden and Plant Care
Authors:

Sune Storm and Stefan Jänicke
Abstract: We present a user-centered design study on interactive garden maintenance through a rule-based system that combines structured plant care data and real-time weather forecasts to support adaptive pruning and watering schedules. We involved end-users and professionals in a participatory design process to shape the task abstractions and interface components. The resulting Growlytics application enables section-based garden configuration, personalized care timelines, and inspiration filtering for ecological and aesthetic planning prioritizing a holistic view of garden maintenance. Our approach balances domain-specific flexibility with visual clarity, offering a lightweight alternative to sensor-driven smart garden systems.
Download
Paper Nr: 71
Title:

Evaluating Large Language Model Integration into Natural Language Interfaces for Visualizations with Available Input Data
Authors:

Adrian Jobst, Daniel Atzberger, Willy Scheibel, Mariia Tytarenko, Jürgen Döllner and Tobias Schreck
Abstract: This paper evaluates an approach for Natural Language Interfaces (NLIs), in which a text-only LLM is provided with (i) the dataset, (ii) an instruction prompt describing context and expected output, and (iii) a visualization specification that formalizes the mapping between data attributes and visual variables. The LLM answers user queries by generating and executing analysis code and returns either textual responses or structured control information that modifies the visualization (e.g., highlighting marks or changing encodings). We conduct three experiments: (1) a quantitative comparison against published multimodal LLM (MLLM) results on an adapted Visualization Literacy Assessment Test (VLAT) benchmark, (2) an extended questionnaire covering six visualization types with both visual-referenced and non-visual questions requiring textual answers, and (3) a set of tasks requiring visual answers via specification edits. Across these evaluations and multiple runs to account for output variability, the proposed integration consistently outperforms prior MLLM-based approaches on VLAT and demonstrates high reliability on both text-answering and visualization-modification tasks, with remaining weaknesses primarily in perceptual/structure judgments such as clusters and anomalies. We release an open-source client–server implementation and evaluation material to support replication and adoption by visualization practitioners.
Download
Paper Nr: 79
Title:

Interactive Visual Analysis of the Danish District Heating Grid
Authors:

Nicklas E. Vrede, Esben A. W. B. Sørensen, Gareth Walsh, Jakob Kusnick and Stefan Jänicke
Abstract: Denmark’s district heating system, serving 66% of households, is a critical component of national energy policy, yet analysis is hindered by fragmented data spanning pricing, fuel sources, and production. We present a visual analytics design study that investigates how integrated visualization can support comparative analysis of district heating networks at national scale. Based on requirements elicited from stakeholders in the Danish energy market, we design a visual analytic workflow that integrates geospatial, temporal, and comparative views to support cross-network analysis. In addition to expert policy analysis, the design explicitly addresses casual and non-expert users by employing intuitive visual metaphors and progressive disclosure to make complex energy data approachable. While building on established visualization components, our contribution lies in the task abstraction, workflow design, and accessibility-oriented design strategies that transform fragmented infrastructure data into actionable insights. An informal evaluation demonstrates how the approach supports policy analysis, stakeholder dialogue, and transparency. Our design workflow is applicable to other distributed energy systems and public infrastructure domains.
Download
Paper Nr: 18
Title:

Using Visualization for Supporting Beginners of a MOBA Game in Improving Their Skills
Authors:

Kanau Tsuchiya and Hiroshi Hosobe
Abstract: League of Legends has been one of the most popular Multiplayer Online Battle Arena (MOBA) games. The complexity of this game makes it difficult for beginners to keep playing. Previous studies analyzed the statistics of the game, such as coordinates and other parameters of players. Some studies estimated how well players performed in matches and which actions led them to win. Another study used storyline visualization that included temporal data and other information, and another analyzed an entire match. However, research on players’ skills in the early stages of the match did not exist. In this paper, we create a visualization system for game beginners to improve their playing skills using a heatmap, a line graph, and a pie chart. We focus on damage trades that happen early in the match. A damage trade is the action of widening a gap in health or resources between the player and the opponent by using the player’s skills and automatic attacks. Beginners often feel it difficult to understand the skill ranges and damage of each character. By reviewing the events, they can improve their skills at the lane phase and play the matches longer to enjoy the game.
Download

Area 3 - Interactive Environments and Human-Computer Interaction

Full Papers
Paper Nr: 45
Title:

Empirical Design Patterns for Resilience: Reducing Cognitive Overload and UI Manipulation in Digital Grid Control Centers
Authors:

Sofie Ostrau, Sascha Kaven, Merlin Engel, Volker Skwarek and Monique Janneck
Abstract: Network control centers are central nodes for the robustness of modern power systems. Particularly in the context of emerging low-voltage (LV) control centers that process decentralized data streams, human-computer interaction (HCI) represents a systemic security vulnerability. While cybersecurity efforts have traditionally focused on technical defenses, deficiencies in the design of user interfaces (UI) create critical security gaps for operational disruptions and targeted UI-based manipulation through cognitive overload and routine blind spots. This study uses a design science research (DSR) methodology (expert interviews N=2, expert workshop N=9) to empirically investigate this domain-specific user experience (UX) security vulnerability. Three structural, domain-specific UX security clusters are identified: (1) cognitive overload due to system complexity, (2) operational incongruity and routine blindness, and (3) abstraction gaps and UI manipulation. Building on these empirical findings, three evidence-based design patterns (context-aware display, anti-habituation feedback, forensic traceability) were developed in a high-fidelity prototype for LV control centers. This work identifies three domain-specific UX-security vulnerabilities in grid control interfaces and derives empirically grounded design patterns to address them. The resulting high-fidelity prototype provides a concrete instantiation of security-aware UX principles, ready for empirical validation through planned A/B testing.
Download
Paper Nr: 47
Title:

Surface vs. Structural Similarity: Aligning Computational Similarity Metrics with Human Judgments
Authors:

Yutaro Sone, Yuki Ninomiya, Kazuhisa Miwa, Yuichiro Sumi, Ryosuke Nakanishi, Eiji Mitsuda, Koji Sato and Tadashi Odashima
Abstract: Many interactive experiences in Human-Computer Interaction (HCI)-such as search, discovery, and recommendation-are powered by computations that “find similar things.” However, the alignment between cognitive theories of similarity-particularly surface versus structural similarity-and the similarity metrics used in systems remains underexplored. This study operationalizes these components for HCI and empirically evaluates their appropriateness. Using books as the domain, we define surface similarity as properties of the cover design (image and title) and structural similarity as properties of the content. Surface similarity is instantiated via cosine similarity on image features from a Vision Transformer (ViT) and cosine similarity between titles using Sentence-BERT (SBERT). Structural similarity is instantiated as cosine similarity between synopses using SBERT. Human similarity judgments were collected through annotation tasks that rated the similarity of cover designs and synopses for book pairs. We find strong correlations between each computational metric and the corresponding human judgments, confirming their alignment. These results indicate that these metrics can serve as proxies for surface and structural components of human similarity judgment. This provides a foundation for HCI interfaces that visualize, adjust, and explain “what is similar” to support discovery, explainable recommendations, and user-steerable exploration from both surface and structural perspectives.
Download
Paper Nr: 54
Title:

CARMEn: CARla-Based, Multi-Agent Immersive Road Environment Simulation
Authors:

Dário Machado, Frederico Pereira, Sérgio Monteiro, Luís Louro, Raul Almeida, Mónica Alexandra Gomes, Elisabete Freitas, Estela Bicho and Emanuel Sousa
Abstract: Immersive driving and pedestrian simulators have become essential tools for research in urban mobility and transportation, particularly in scenarios involving driver-pedestrian coexistence. By allowing both types of agents to engage with Virtual Environments (VEs), these simulators enable the study of complex behaviours under controlled, repeatable and safe conditions. Recent advances in networked simulations, as well as the the growing availability of both Extended Reality (XR) Head-Mounted Displays and user-friendly acoustic propagation engines, have significantly lowered the barriers to deploying immersive simulations with Multi-Agent, XR and realistic spatial audio capabilities. In this context, we introduce CARMEn, an open-source immersive road environment simulation which integrates these three capabilities into an unified, lower cost framework, built on the CARLA driving simulator and Unreal Engine. CARMEn’s client-server architecture allows both drivers and pedestrians to coexist within the same VE. The framework integrates Augmented Virtuality to merge real-world elements such as the steering wheel and driver’s hands into a high-fidelity virtual automobile cockpit. Finally, it renders binaural spatial audio for each agent’s point of view. We describe the system architecture, software and hardware components, the current state of development, and future improvements. Additionally, we evaluate simulation latency and discuss the potential applications of CARMEn for multi-user, immersive studies in transportation and mobility research.
Download
Paper Nr: 66
Title:

GAINS: Gaussian-Adaptive Interaction for Nonlinear Scaling in VR Reach
Authors:

Wataru Kimura and Tomokazu Ishikawa
Abstract: Virtual arm extension techniques are essential for enabling efficient object manipulation in Virtual Reality (VR) environments, where the limited physical space constrains natural interaction with distant virtual objects. The classic Go-Go technique and its variants have primarily focused on reaching efficiency but often compromise precision when manipulating objects at extended distances due to amplified hand tremors and unintended movements. We propose GAINS, a novel arm extension technique that employs a mixed Gaussian field to dynamically adjust manipulation gain based on proximity to target objects, ensuring both path independence and precise manipulation near targets. In a user study with 17 participants, we compared GAINS against three existing techniques: Go-Go, PRISM, and Potential (a path-dependent variant). Our results show that while Go-Go achieved the fastest task completion time (18.60s) and second highest usability scores (SUS: 71.94), GAINS demonstrated competitive performance in mistouch reduction (1.40 mistouches) compared to the path-dependent Potential (1.34 mistouches), while maintaining superior path independence. GAINS achieved completion times of 19.01s with moderate NASA-TLX scores (51.24) and SUS scores (71.11). These findings highlight the fundamental trade-off between reach efficiency and manipulation precision in VR interfaces, suggesting that optimal technique selection depends on task-specific requirements for accuracy versus speed.
Download
Short Papers
Paper Nr: 13
Title:

The Effects of Repeated iVR Exposure on Performance, Motivation, and Anxiety in English Listening
Authors:

Bruno Peixoto, Luciana Cabral Pereira Bessa, Guilherme Gonçalves, Maximino Bessa and Miguel Melo
Abstract: As Immersive Virtual Reality (iVR) shows promise in education, most research relies on single-session interventions, leaving the sustained impact of its systematic use largely unexplored. This exploratory, semester-long study begins to address this gap by investigating the effects of a repeated iVR intervention on university students’ motivation, anxiety, and performance in English listening comprehension. Over one semester, 45 undergraduate students participated in traditional audio-based listening exercises alongside two custom-designed iVR tasks. Motivation and anxiety were measured at three time points, while performance, presence and cyber-sickness were assessed following each iVR task. Initial findings indicate that while iVR did not significantly alter most motivational orientations, a notable decrease in the perceived utility of the tasks was observed. Furthermore, students performed significantly better on traditional listening tasks compared to the iVR exercises, suggesting that immersive environments may introduce additional cognitive load and distraction. These preliminary results suggest that while iVR is an engaging tool, its pedagogical benefits are contingent on careful instructional design that manages cognitive demands and aligns with learner goals.
Download
Paper Nr: 17
Title:

Exploring How Social Context Aware NPCs Change User Presence and Engagement in VR Experiences
Authors:

Anders Lucassen Lund, Jakob Krogsgaard Pedersen, Markus Birch Flensborg, Rikke Bragh Jensen, Tze Huo Gucci Ho and Ivan Nikolov
Abstract: Enhancing immersion and believability in VR experiences can improve learning, adherence to instructions, and overall user experience. An often overlooked factor is contextual social norm interaction between NPCs and users. NPCs are typically used for exposition, limiting immersion and presence. This paper explores how context-aware social interactions impact user experience and presence. We created an office scenario with a task list and two NPC versions: one that responds to user actions, context, and proximity, and one that does not. We also propose a custom presence questionnaire for detecting how users’ behavior, immersion and presence change when interacting with the two version. Results show that the context-aware NPCs make users more hesitant to explore and interact, leading to a quicker sense of frustration, but also increases social presence and the feeling of being there. The paper is an explorative study into developing more reactive NPCs for games that have internal moods that guide how they emote, react and communicate with users, based on users’ actions.
Download
Paper Nr: 23
Title:

Experiential Perception Data Collection Using Augmented Reality
Authors:

Marlene Huber, Hannes Kaufmann and Milena Vuckovic
Abstract: This paper presents the design and implementation of a portable, mobile-based AR prototype that employs situated visualizations to support participatory, citizen-driven geospatial data mapping. Our approach focuses on collecting experiential perception data in real urban contexts using a radial menu interface, with a specific emphasis on personal thermal sensations and their influence on the perceived thermal quality of public spaces. To evaluate the usability and practicality of the prototype, we conducted a user study in a real-world urban environment with 21 participants and employed a structured questionnaire using a 5-point Likert scale to measure users’ subjective satisfaction with the system’s usability. The results demonstrate the system’s high usability and indicate that radial menus are an effective method for multivariate data collection.
Download
Paper Nr: 33
Title:

Perceptual Evaluation of Physically-Based Sound Propagation in Virtual Environments
Authors:

Fredrik Svensson, Veronica Sundstedt, Yan Hu and Andreas Rossholm
Abstract: The development of spatial audio in virtual environments (VEs) has made significant progress, particularly with the introduction of physics-based sound propagation that incorporates material properties. However, research remains limited on how users perceive these audio enhancements in virtual reality (VR), especially regarding immersion. This paper investigates how material properties influence the user experience of spatial audio in VR, focusing on perceived realism, presence, and preference. A VR application was developed in Unreal Engine 5.5 using the Steam Audio plugin to simulate sound propagation across four common indoor materials: concrete, glass, carpet, and plaster. Each material was tested in three acoustic configurations. A fifth environment combining all materials was also tested. Participants freely explored the virtual space and rated each material configuration on audio realism on a 0–100 Likert scale. Three questionnaires were administered: pre-experiment, post-experiment, and the Simulator Sickness Questionnaire (SSQ). Results showed that while participants could distinguish between materials, they struggled to assess acoustic configuration differences. Sound propagation was generally perceived as natural and realistic, with materials rated between ‘Fair’ and ‘Good’. Findings suggest that physics-based sound propagation enhances immersion in VR, though user familiarity with materials may significantly influence perception.
Download
Paper Nr: 35
Title:

Integrating Chatbots into Virtual Reality with Generative AI: A Brief Review
Authors:

Omer Emin Cinar, Karen Rafferty and Hui Wang
Abstract: The convergence of Virtual Reality (VR), Generative Artificial Intelligence (AI), and chatbot technologies represents a transformative paradigm in Human-Computer Interaction (HCI). This review examines the integration of AI-powered chatbots within virtual environments, exploring how generative AI enhances immersive conversational experiences. Through analysis of current implementations across education and healthcare domains, key technological advances, implementation challenges, and design considerations are identified. The findings indicate that Large Language Models (LLMs) integrated with VR create more natural, context-aware interactions that significantly improve user engagement and learning outcomes. This paper provides an overview of the current state and future directions for AI-powered conversational agents in VR environments.
Download
Paper Nr: 39
Title:

Phenomenon-Based Design for Mixed Reality: Sustaining Real-Virtual Relationships through Physical Phenomena
Authors:

Toshiro Kashiwagi and Kumiyo Nakakoji
Abstract: The real world always serves as the foundation upon which virtual objects are superimposed in optical see-through MR environments such as those using the HoloLens 2. Users directly perceive real light, texture, and motion while also perceiving virtual information. Naturally, physical phenomena in the real world are not static backgrounds but dynamic elements that continuously change over time and form the preconditions of MR experiences. We have developed two MR tools, DITTO-Balance and DITTO-LightFilter, to explore how virtual objects can participate within phenomena that already occur in the real world. DITTO-Balance allows a virtual weight to influence the equilibrium of a physical balance, while DITTO-LightFilter enables a color change to appear within the real behavior of light by manipulating a virtual object.
Download
Paper Nr: 52
Title:

Yahp!: Yet Another Haptic Probe
Authors:

Martín Tau, Juan Lucas Gutierrez, Andres Rodriguez and Federico Balaguer
Abstract: Characterizing the relationship between vibratory stimuli and user responses is a complex challenge due to varying skin sensitivity across body areas and the onset of stimulus saturation. Achieving an optimal balance between body location, actuator types, and haptic cues is often a demanding and error-prone process. This paper presents Yahp! (Yet Another Haptic Probe), an open-source tool developed through a collaboration between the National University of La Plata and Stream S.A. to systematically design and execute user perception tests for industrial haptic alerts. Yahp! facilitates the evaluation of actuator settings and body locations through a modular architecture consisting of formal experiment definitions, a results database, a mobile trial director, and a generic haptic device utilizing a low-level messaging protocol. To demonstrate the tool’s utility, we present two experiments focusing on haptic bracelets and sleeves. Our preliminary results indicate that while 10% vibration intensity is consistently below the detection threshold, higher intensities are reliably perceived within an average of 3.5 seconds. Furthermore, the studies revealed significant detectability asymmetries during intensity transitions and confirmed the impact of sensory saturation on cue recognition. These findings suggest that Yahp! is an effective platform for defining the symbolic language of haptic interfaces in real-world applications..
Download
Paper Nr: 56
Title:

User Experience of PAL-HAND.Q, a Pneumatic Haptic Device for Finger-Level Gaming Interaction
Authors:

Fabrizio Sulpizio, Ciro Annicchiarico, Giovanni Colucci, Simone Duretto, Francesco Strada, Giuseppe Quaglia and Andrea Bottino
Abstract: This study investigates the usability and user experience of PAL-HAND.Q, a handheld haptic device designed to provide independent haptic feedback on each finger through an integrated electro-pneumatic system. The device features five soft pneumatic membranes-one per finger-enabling vibrotactile stimulation and variable stiffness feedback. We conducted a usability study using two games that leverage the device’s key features: Tile Game, which emphasizes timed actions and finger coordination, and Airplane Game, which integrates finger pressing with device orientation control. Twenty-five participants tested the system and completed questionnaires on usability, workload, game experience, haptic experience, and comfort. The results indicate good usability, moderate perceived workload, and engaging interactions. Notably, the device demonstrated better performance in continuous control tasks compared to time-pressured precision tasks, suggesting its suitability for applications requiring sustained, smooth finger-level interaction. Overall, the findings demonstrate PAL-HAND.Q’s effectiveness for finger-level gaming interaction and point to its potential applicability in other domains requiring portable, independent finger-level haptic control, such as virtual and augmented reality, rehabilitation, and interactive training systems.
Download
Paper Nr: 64
Title:

Real-Time Walking-in-Place Recognition Using Multi-Point Inertial Sensors and a Lightweight CNN
Authors:

Yu Takahashi and Tomokazu Ishikawa
Abstract: Motion recognition based on Walking-in-Place (WIP) enables users to walk naturally within virtual reality (VR) without physical translation and helps mitigate VR sickness that commonly arises with conventional controller-based locomotion. However, existing WIP techniques face challenges for practical deployment, including low recognition accuracy in real-world settings and inference latencies long enough to compromise real-time operation. We propose a new VR locomotion method based on recognizing stepping motions using inertial sensors. Our method uses six inertial sensors to capture full-body motion data and adopts an efficient single-stream convolutional neural network (CNN) that recognizes actions within a short 0.64s window. To address coordinate drift (accumulated error) during prolonged sensor use, instead of feeding raw position estimates, we use velocities and accelerations as input features, achieving high recognition accuracy while maintaining real-time performance. In an evaluation using a custom runner game, we show that while keyboard control demonstrated superior task performance (97.8% vs. 49.1% success rate), the proposed WIP method achieved significantly higher immersion and enjoyment ratings, with 12 of 16 participants preferring it for entertainment applications.
Download
Paper Nr: 22
Title:

Do Our Design Principles Ever Really Change, or Do We Just Change the Methods of Achieving Them?
Authors:

Anastasiia Satarenko
Abstract: Human–computer interaction has long been guided by principles such as system feedback, directness, and usability, most clearly articulated in Direct Manipulation Interfaces (DMIs). With the emergence of adaptive, AI-driven systems, new interaction paradigms raise the question of whether these foundations still hold or must be redefined. In this paper, we introduce Infinite Interfaces (IIs), a novel paradigm that integrates natural language processing, contextual awareness, and multimodal outputs into a single, adaptive entry point. To examine whether traditional principles extend to this new context, we conducted a moderated usability study with 10 participants, testing an II prototype on tasks ranging from simple actions to complex, multi-step problem-solving. The findings reveal that while IIs introduce new requirements-such as semantic alignment, personalization, progressive disclosure of functionality, and visually distinguishable multimodal outputs-they remain grounded in the same core principles as DMIs. Rather than replacing established foundations, IIs reinterpret and extend them. This work contributes to HCI theory by demonstrating HCI Researchers and UI/UX Designers that design principles remain stable across technological shifts, with innovation emerging in the methods used to achieve them.
Download
Paper Nr: 29
Title:

A Rigorous Protocol for Evaluating a Virtual Reality Anti-Bullying Simulation for Youth: A Proposed Study
Authors:

Tanvir Irfan Chowdhury
Abstract: Bullying remains a critical public health issue with profound consequences for youth, yet traditional prevention strategies often fail to engage students actively. This paper presents the design and a rigorous experimental protocol for evaluating a Virtual Reality Anti-bullying Simulation (VAS) aimed at educating children ages 13–18. The proposed study employs a between-subjects design (planned N = 100) to compare the immersive VAS experience against a non-immersive tablet-based control. Moving beyond simple knowledge tests, the protocol utilizes a multi-dimensional evaluation framework that includes the Victim Concern Scale to measure attitudinal change, an Information Recall Questionnaire (IRQ) for knowledge retention, and standardized measures of presence and embodiment. We hypothesize that the VR group will demonstrate: (1) stronger provictim attitudes and (2) superior information recall than the control group, and that (3) these outcomes will be positively correlated with the psychological mechanisms of presence and embodiment. This work contributes a robust methodological framework for assessing immersive learning tools for sensitive social issues.
Paper Nr: 42
Title:

Redirection through Shifting Walls and Implausible Corridors
Authors:

Mathieu Lutfallah, Yannis Müri, Christian Popescu and Andreas Kunz
Abstract: This work introduces two redirection techniques based on virtual environment (VE) manipulation. The first alters the orientation of the VE relative to the physical space using seamless teleportation in a corridor that allows users to skip certain circular segments. The second shifts the position of the VE by moving walls when the user is not looking. Both techniques were evaluated in user studies with 21 and 22 participants, respectively, under varying degrees of manipulation to assess detection thresholds and user perception. For the teleportation, half of the participants did not detect any manipulation; those who did either observed a change in corridor shape while assuming consistent orientation, or noticed a change in the spatial layout. The shifting walls enabled displacements up to 28.4% of the room’s side length without being noticed. Both techniques showed low disturbances and cognitive load, highlighting their potential for future use.
Download
Paper Nr: 43
Title:

AnitoTracer: An Interactive Ray Tracer for Designing 3D Ray-Traced Scenes
Authors:

Neil Patrick Del Gallego, Kate Nicole Young, Shane Laurenze Cablayan, Marcus Rene Levin Leocario, Zachary Gadjiel Breinard Que and Andre Vito Valdecantos
Abstract: Ray tracing (RT) is a rendering technique used in video games and computer graphics to produce photorealistic scenes. Existing 3D modelling software and game engines, like Unity, Unreal, and Blender, now support real-time ray tracing to facilitate the creation of virtual worlds and games with RT. However, enabling these features often requires users to navigate multiple editor settings, install specific packages or plugins, and alter the default rendering pipelines. Such a method can introduce a steep learning curve, interrupt the creative workflow, and make it less accessible to developers who are not yet familiar with designing 3D assets compatible with RT. In this study, we developed AnitoTracer, an interactive ray tracing engine with scene editor features that allow users to composite 3D scenes with ray tracing already enabled. Its architecture is specifically designed to provide a cohesive, real-time, RT scene editor for users without the complex setup process found in other 3D engines. We implemented AnitoTracer using C++, open-source libraries, and Vulkan's ray tracing API as our renderer. We conducted a usability test where 15 participants tested the RT engine. AnitoTracer has a Standard Usability (SUS) score of 68.33, indicating that the engine is usable in its current form and provides a solid foundation for further development. There is potential for improvement in optimization and stability, as well as the addition of more 3D-modelling user interactions, such as keyboard shortcuts, undo-redo commands, and pixel-perfect object picking. AnitoTracer is open source, and programmers can freely extend the capabilities of the prototype to suit their needs. AnitoTracer's source code can be accessed at this link: https://dlsu-game-lab.github.io/AnitoTracer/.
Download
Paper Nr: 50
Title:

AR and GenAI: Is Generative Artificial Intelligence the Answer to Rendering Challenges in Augmented Reality?
Authors:

Claus B. Madsen
Abstract: The paper examines Generative Artificial Intelligence (GenAI) from a perspective of its ability to do visually realistic Augmented Reality (AR). Through examples we show how impressively well GenAI can render virtual/augmented objects into still images, and addressing in turn each of the three main AR challenges: camera tracking, illumination consistency, and occlusion handling. The paper then discusses the strengths and weaknesses of GenAI in this context, and points out directions for future research. One such direction is temporal consistency. The message of the paper is that while GenAI can certainly generate extremely convincing images, far exceeding what would be possible with a traditional model-based approach, we are still very far from being able to use GenAI as a rendering approach for functional AR applications.
Download
Paper Nr: 58
Title:

UsherXR: Placing Remote Students in Their Classroom Seats as 2D Video Sprites
Authors:

Franklin Liu, Siyu Huang and Voicu Popescu
Abstract: This paper introduces UsherXR, an extended reality (XR) system that integrates both remote and local students into the same seat with little to no latency. The instructor wears an XR headset running in passthrough mode that renders remote students as live video sprites virtually placed into the empty seats of the classroom. UsherXR achieves scalability with the number of students by integrating the individual remote student video feeds into a video atlas on an edge server, which is then sent to the headset of the instructor. UsherXR was evaluated with good results in a controlled IRB-approved user study that investigated the effectiveness of UsherXR from the remote and local student perspectives.
Download
Paper Nr: 65
Title:

Effects of Music Tempo on Vection Perception in Virtual Reality Environments
Authors:

Takuto Suganuma and Tomokazu Ishikawa
Abstract: Vection, the illusion of self-motion induced by sensory stimulation, is one of the critical phenomena that enhance presence in virtual reality (VR) environments. While previous studies have examined the effects of spatial characteristics of visual and auditory stimuli on vection, research focusing on music tempo (beats per minute, BPM) remains extremely limited. In this study, we empirically investigated how music stimuli with different tempos (nine levels from 40 to 200 BPM) affect vection perception in a VR environment using a headmounted display (HMD). Experimental results with 15 participants revealed a significant positive correlation between BPM and vection intensity (r = 0.255, p < 0.000001), indicating that faster tempos enhance vection intensity. Furthermore, a significant negative correlation was found between BPM and vection onset time (r = -0.158, p = 0.0003), demonstrating that faster tempos induce vection more rapidly. Comprehensive evaluation identified 140-160 BPM as the most effective tempo range for inducing vection. This optimal range is significantly faster than the natural tempo range (85-120 BPM) reported in music cognition research, suggesting unique temporal characteristics in audiovisual integration. These findings indicate that appropriate music tempo configuration in VR content design can effectively enhance presence and immersion.
Download

Area 4 - Scientific Visualization

Full Papers
Paper Nr: 78
Title:

Holoview: An Immersive AR System for Interactive Anatomy Visualization and Learning
Authors:

Ojaswa Sharma and Anshul Goswami
Abstract: We present Holoview, an augmented reality (AR) system designed to support immersive and interactive learning of human anatomy. Holoview enables users to dynamically explore volumetric anatomical data through intuitive hand gestures in a 3D AR environment, allowing inspection of individual organs and cross-sectional views via clipping and bioscope features. The system adopts a lightweight client–server architecture optimized for real-time performance on the HoloLens through hybrid and foveated rendering. Our formative user study with participants possessing foundational anatomical knowledge demonstrated improvements in task-specific understanding and reported high engagement with gesture-based interactions. The system was perceived as engaging and intuitive, particularly for organ selection and cross-sectional exploration, with low cognitive load and increasing ease of use over time. These findings highlight Holoview’s potential as an evaluation step toward immersive, user-centered AR systems for anatomy education.
Download
Short Papers
Paper Nr: 11
Title:

Bubble Behavior Analysis Considering Kinetic Energy
Authors:

Nobuhiko Mukai, Koki Tomita, Takuya Tajima and Youngha Chang
Abstract: Visualization of bubble behavior is one of the most challenging issues since bubbles are composed of two fluids. One is liquid and the other is air, while the density ratio of the two is about 1,000:1. Air, having a low density, is pushed and compressed by liquid that has a high density. A two-path solution, which calculates the governing equations separately, is often used, generating a gap on the boundary between the two fluids. Another one is a one-path that calculates the equations simultaneously for both fluids with the condition that the density ratio is low, which cannot simulate the bubble behavior precisely. In general, liquid has high density and low kinetic energy, while air has low density and high kinetic energy. In this paper, we solve the Navier-Stokes equations simultaneously for both fluids with a high-density ratio by considering kinetic energy for both fluids and visualize the behavior of a bubble.
Download
Paper Nr: 36
Title:

ANIA: 3D Asset Generation Engine
Authors:

Mariana Reis, João Quintas and Paulo Menezes
Abstract: This paper presents ANIA, a model-agnostic 3D asset generation engine that converts one to four RGB images into optimized, textured 3D meshes using generative AI. Rather than proposing a new generative backbone, ANIA focuses on the engineering and integration of existing geometry and texture generation models into a coherent, reproducible, and self-hostable pipeline. The system comprises five stages-input preprocessing, 3D mesh generation, mesh optimization, multi-view image generation, and UV-mapped texture synthesis (baking and inpainting)-implemented in a modular, node-driven architecture. By combining diffusion-based and attention-based models with classical mesh decimation and UV layout techniques, ANIA produces game-ready assets within minutes instead of days. We evaluate runtime and qualitative output across single- and multi-view configurations and compare results against manually created baselines and closed-source systems. Finally, we discuss current limitations, including animation-ready topology, lighting consistency, and residual texture artifacts, and outline a roadmap toward production readiness, encompassing quad-dominant retopology, color-consistent multi-view refinement, and animation support.
Download

Area 5 - Animation and Simulation

Full Papers
Paper Nr: 27
Title:

Pinch Joints: Real-Time Simulation of 2D Cable Pinching
Authors:

Nicolas Elvstål Ångnell, Prashant Goswami, Veronica Sundstedt and Yan Hu
Abstract: Cable and rope simulations play an important role across diverse domains, including interactive graphics and gaming. Achieving real-time performance remains challenging, as most existing approaches either exhibit poor computational scalability or compromise essential physical dynamics. Moreover, critical effects such as the pinching of cables between objects are typically not captured by current methods. To address these limitations and assess the importance of pinching, this paper introduces an extension to Cable Joints (Müller et al., 2018) utilizing collision detection and contact constraints to enforce the width of the cable when pinched in a two-dimensional setting. The resulting cable model shows promise for applications with strict performance limits. Performance evaluations demonstrate that the proposed extension maintains suitability for real-time applications. Furthermore, results of a perceptual evaluation indicate a strong preference for the enhanced cable dynamics introduced by the novel pinching method.
Download
Short Papers
Paper Nr: 57
Title:

Combining Large Language Models with Procedural Grammars for Scenario Generation in Driving Simulations
Authors:

Nelson Bilber Rodrigues, António Coelho and Rosaldo J. F. Rossetti
Abstract: Scenario-based simulation is a crucial tool for assessing and validating systems across a wide range of applications, from traffic engineering studies to educational purposes. However, creating detailed, realistic simulation scenarios requires significant manual customisation, which limits the diversity and expressiveness of the scenes. This paper investigates how Large Language Models (LLMs) can be combined with procedural content generation (PCG) techniques to transform narrative descriptions into executable driving simulations. It explores how narrative descriptions can be transformed into structured geospatial representations by using LLMs for generating road topology layouts. Also, how to extract patterns and parametrizations for representing dynamic traffic manoeuvrers. Two case studies were implemented to reproduce a cut-in manoeuvre: the case study uses LLM to fill meta-templates based on OpenSCENARIO descriptions and executes them in esmini simulator. The second approach integrates Retrieval-Augmented Generation (RAG) with a grammar-based procedural representation to transform natural-language instructions into a symbolic behavioural language executable in the CARLA simulator. This integrated approach promises to reduce manual intervention, facilitates the rapid exploration of diverse case studies, and contribute to more efficient and scalable scenario generation for various simulation applications.
Download
Paper Nr: 63
Title:

Parse, Plan and Move: Text-Driven Multi-Task Diffusion for Human-Scene Interaction Generation
Authors:

Jia Chen and Yingying Wang
Abstract: Generating human-scene interaction (HSI) has broad applications in virtual reality, augmented reality and games. Text instructions provide a friendly and accessible interface for users to specify their high-level scene interaction requirements. However, it is challenging to build a generative HSI model that understands both users’ arbitrary textual instructions and dense geometric scene constraints. In this work, we propose a framework that parses multi-task instructions, plans and implements HSI motions in three stages. First, the task parser parses long complex user instructions, and decomposes them into atomic sub-tasks. Then, the planner takes sub-task instructions together with 2D maps of the scene as input, and outputs sub-task plans and sparse motion guidances for interactions of different types. Lastly, based on the guidances and the 3D scene meshes, the implementer generates diverse HSI motion segments through diffusion for corresponding sub-tasks. Our framework is fully automatic, and handles multi-modal input data by integrating foundational large language models (LLMs) for processing texts, and motion diffusion models (MDM) for understanding the 3D scene and motion constraints. We demonstrate initial qualitative results generated from this framework, showing great promise in text-driven HSI generation tasks.
Download