EG 2025 - Short Papers

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/3607138

Browse

Now showing 1 - 20 of 27

Non-linear, Team-based VR Training for Cardiac Arrest Care with enhanced CRM Toolkit
(The Eurographics Association, 2025) Kentros, Mike; Kamarianakis, Manos; Cole, Michael; Popov, Vitaliy; Protopsaltis, Antonis; Papagiannakis, George; Ceylan, Duygu; Li, Tzu-Mao
This paper introduces iREACT, a novel VR simulation addressing key limitations in traditional cardiac arrest (CA) training. Conventional methods struggle to replicate the dynamic nature of real CA events, hindering Crew Resource Management (CRM) skill development. iREACT provides a non-linear, collaborative environment where teams respond to changing patient states, mirroring real CA complexities. By capturing multi-modal data (user actions, cognitive load, visual gaze) and offering real-time and post-session feedback, iREACT enhances CRM assessment beyond traditional methods. A formative evaluation with medical experts underscores its usability and educational value, with potential applications in other high-stakes training scenarios to improve teamwork, communication, and decision-making.
Multi-Objective Packing of 3D Objects into Arbitrary Containers
(The Eurographics Association, 2025) Meißenhelter, Hermann; Weller, Rene; Zachmann, Gabriel; Ceylan, Duygu; Li, Tzu-Mao
Packing problems arise in numerous real-world applications and often take diverse forms. We focus on the relatively underexplored task of packing a set of arbitrary 3D objects-drawn from a predefined distribution-into a single arbitrary 3D container. We simultaneously optimize two potentially conflicting objectives: maximizing the packed volume and maintaining sufficient spacing among objects of the same type to prevent clustering. We present an algorithm to compute solutions to this challenging problem heuristically. Our approach is a flexible two-tier pipeline that computes and refines an initial arrangement. Our results confirm that this approach achieves dense packings across various objects and container shapes.
PartFull: A Hybrid Method for Part-Aware 3D Object Reconstruction from Sparse Views
(The Eurographics Association, 2025) Yao, Grekou; Mavromatis, Sébastien; Mari, Jean-Luc; Ceylan, Duygu; Li, Tzu-Mao
Recent advancements in 3D object reconstruction have been significantly enhanced by generative models; however, challenges remain when detailed 3D shapes are reconstructed from limited, sparse views. Traditional methods often require multiple input views and known camera poses, whereas newer approaches that leverage diffusion models from single images encounter realworld data limitations. In response, we propose ''PartFull'', a novel framework for part-aware 3D reconstruction using a hybrid approach. ''PartFull'' generates realistic 3D models from sparse RGB images by combining implicit and explicit representations to optimize surface reconstruction. Starting with sketch-based 3D models from individual views, these models are fused into a coherent object. Our pipeline incorporates a pretrained latent space for part-aware implicit representations and a deformable grid for feature volume construction and surface optimization. PartFull's joint optimization of surface geometry, topology, and implicit part segmentation constitutes a new approach to addressing the challenges of 3D reconstruction from sparse views.
Double QuickCurve: revisiting 3-axis non-planar 3D printing
(The Eurographics Association, 2025) Ottonello, Emilio; Hugron, Pierre-Alexandre; Parmiggiani, Alberto; Lefebvre, Sylvain; Ceylan, Duygu; Li, Tzu-Mao
Additive manufacturing builds physical objects by accumulating layers of solidified material. This is typically done with planar layers. Fused filament printers however have the capability to extrude material along 3D curves, leading to the idea of depositing in a non-planar fashion. In this paper we introduce a novel algorithm for this purpose, targeting simplicity, robustness and efficiency. Our method interpolates curved slicing surfaces between two top and bottom slicing surfaces, optimized to align with the object curvatures. These slicing surfaces are intersected with the input model to extract non-planar layers and curved deposition trajectories. We further orient trajectories according to the object's curvatures, improving deposition.
3D Garments: Reconstructing Topologically Correct Geometry and High-Quality Texture from Two Garment Images
(The Eurographics Association, 2025) Heße, Lisa; Yadav, Sunil; Ceylan, Duygu; Li, Tzu-Mao
We present a fully integrated pipeline for generating topologically correct 3D meshes and high-fidelity textures of fashion garments. Our geometry reconstruction module takes two input images and employs a semi-signed distance field representation with shifted generalized winding numbers in a deep-learning framework to produce accurate, non-watertight meshes. To create realistic, high-resolution textures (up to 4K) that closely match the input, we combine diffusion-based inpainting with a differentiable renderer, further enhancing the quality through normal-guided projection to minimize projection distortions in the texture image. Our results demonstrate both precise geometry and richly detailed textures. In addition, we are making a portion of our high-quality training dataset publicly available, consisting of 250 lower-garment triangulated meshes with 4K textures.
NoiseGS: Boosting 3D Gaussian Splatting with Positional Noise for Large-Scale Scene Rendering
(The Eurographics Association, 2025) Kweon, Minseong; Cheng, Kai; Chen, Xuejin; Park, Jinsun; Ceylan, Duygu; Li, Tzu-Mao
3D Gaussian Splatting (3DGS) efficiently renders 3D spaces by adaptively densifying anisotropic Gaussians from initial points. However, in complex scenes such as city-scale environments, large Gaussians often overlap with high-frequency regions rich in edges and fine details. In these areas, conflicting per-pixel gradient directions cause gradient cancellation, reducing the overall gradient magnitude and potentially causing Gaussians to remain trapped in suboptimal positions even after densification. To address this, we propose NoiseGS, a novel approach that integrates randomized noise injection into 3DGS, guiding suboptimal Gaussians selected for densification toward more optimal positions. In addition, to mitigate the instability caused by oversized Gaussians, we introduce an ℓp-penalization on the scale of Gaussians. Our method integrates seamlessly with existing heuristicbased optimization and demonstrates strong generalization in reconstructing complex scenes such as MatrixCity and Building.
Automated Skeleton Transformations on 3D Tree Models Captured from an RGB Video
(The Eurographics Association, 2025) Michels, Joren; Moonen, Steven; GÜNEY, ENES; Temsamani, Abdellatif Bey; Michiels, Nick; Ceylan, Duygu; Li, Tzu-Mao
A lot of work has been done surrounding the generation of realistically looking 3D models of trees. In most cases, L-systems are used to create variations of specific trees from a set of rules. While achieving good results, these techniques require knowledge of the structure of the tree to construct generative rules. We propose a system that can create variations of trees captured by a single RGB video. Using our method, plausible variations can be created without needing prior knowledge of the specific type of tree. This results in a fast and cost-efficient way to generate trees that resemble their real-life counterparts.
LabanLab: An Interactive Choreographical System with Labanotation-Motion Preview
(The Eurographics Association, 2025) Yan, Zhe; Yu, Borou; Wang, Zeyu; Ceylan, Duygu; Li, Tzu-Mao
This paper introduces LabanLab, a novel choreography system that facilitates the creation of dance notation with motion preview. LabanLab features an interactive interface for creating Labanotation staff coupled with visualization of corresponding movements. Leveraging large language models (LLMs) and text-to-motion frameworks, LabanLab translates symbolic notation into natural language descriptions to generate lifelike character animations. As the first web-based Labanotation editor with motion synthesis capabilities, LabanLab makes Labanotation an input modality for multitrack human motion generation, empowering choreographers with practical tools and inviting novices to explore dance notation interactively.
Audio-aided Character Control for Inertial Measurement Tracking
(The Eurographics Association, 2025) Jang, Hojun; Bae, Jinseok; Kim, Young Min; Ceylan, Duygu; Li, Tzu-Mao
Physics-based character control generates realistic motion dynamics by leveraging kinematic priors from large-scale data within a simulation engine. The simulated motion respects physical plausibility, while dynamic cues like contacts and forces guide compelling human-scene interaction. However, leveraging audio cues, which can capture physical contacts in a costeffective way, has been less explored in animating human motions. In this work, we demonstrate that audio inputs can enhance accuracy in predicting footsteps and capturing human locomotion dynamics. Experiments validate that audio-aided control from sparse observations (e.g., an IMU sensor on a VR headset) enhances the prediction accuracy of contact dynamics and motion tracking, offering a practical auxiliary signal for robotics, gaming, and virtual environments.
Lightweight Morphology-Aware Encoding for Motion Learning
(The Eurographics Association, 2025) Wu, Ziyu; Michel, Thomas; Rohmer, Damien; Ceylan, Duygu; Li, Tzu-Mao
We present a lightweight method for encoding, learning, and predicting 3D rigged character motion sequences that consider both the character's pose and morphology. Specifically, we introduce an enhanced skeletal embedding that extends the standard skeletal representation by incorporating the radius of proxy cylinders, which conveys geometric information about the character's morphology at each joint. This additional geometric data is represented using compact tokens designed to work seamlessly with transformer architectures. This simple yet effective representation demonstrated through three distinct tokenization strategies, maintains the efficiency of skeletal-based representations while enhancing the accuracy of motion sequence predictions across diverse morphologies. Notably, our method achieves these results despite being trained on a limited dataset, showcasing its potential for applications with scarce animation data.
Implicit Shape Avatar Generalization across Pose and Identity
(The Eurographics Association, 2025) Loranchet, Guillaume; Hellier, Pierre; Schnitzler, Francois; Boukhayma, Adnane; Regateiro, Joao; Multon, Franck; Ceylan, Duygu; Li, Tzu-Mao
The creation of realistic animated avatars has become a hot-topic in both academia and the creative industry. Recent advancements in deep learning and implicit representations have opened new research avenues, particularly in enhancing avatar details with lightweight models. This paper introduces an improvement over the state-of-the-art implicit Fast-SNARF method to permit generalization to novel motions and shape identities. Fast-SNARF trains two networks: an occupancy network to predict the shape of a character in canonical space, and a Linear Blend Skinning network to deform it into arbitrary poses. However, it requires a separated model for each subject. We extend this work by conditioning both networks on an identity parameter, enabling a single model to generalize across multiple identities, without increasing the model's size, compared to Fast-SNARF.
Parallel Dense-Geometry-Format Topology Decompression
(The Eurographics Association, 2025) Meyer, Quirin; Barczak, Joshua; Reitter, Sander; Benthin, Carsten; Ceylan, Duygu; Li, Tzu-Mao
Dense Geometry Format (DGF) [BBM24] is a hardware-friendly representation for compressed triangle meshes specifically designed to support GPU hardware ray tracing. It decomposes a mesh into meshlets, i.e., small meshes with up to 64 positions, triangles, primitive indices, and opacity values, in a 128-byte block. However, accessing a triangle requires a slow sequential decompression algorithm with O(T) steps, where T is the number of triangles in a DGF block. We propose a novel parallel algorithm with O(logT) steps for arbitrary T. For DGF, where T ≤ 64, we transform our algorithm to allow O(1) access. We believe that our algorithm is suitable for hardware implementations. With our algorithm, a custom intersection shader outperforms the existing serial decompression method. Further, our mesh shader implementation achieves competitive rasterization performance with the vertex pipeline. Finally, we show how our method may parallelize other topology decompression schemes.
Controlled Image Variability via Diffusion Processes
(The Eurographics Association, 2025) Zhu, Yueze; Mitra, Niloy J.; Ceylan, Duygu; Li, Tzu-Mao
Diffusion models have shown remarkable abilities in generating realistic images. Unfortunately, diffusion processes do not directly produce diverse samples. Recent work has addressed this problem by applying a joint-particle time-evolving potential force that encourages varied and distinct generations. However, such a method focuses on improving the diversity across any batch of generation rather than producing variations of a specific sample. In this paper, we propose a method for creating subtle variations of a single (generated) image - specifically, we propose Single Sample Refinement, a simple and training-free method to improve the diversity of one specific sample at different levels of variability. This mode is useful for creative content generation, allowing users to explore controlled variations without sacrificing the identity of the main objects.
Pixels2Points: Fusing 2D and 3D Features for Facial Skin Segmentation
(The Eurographics Association, 2025) Chen, Victoria Yue; Wang, Daoye; Garbin, Stephan; Bednarik, Jan; Winberg, Sebastian; Bolkart, Timo; Beeler, Thabo; Ceylan, Duygu; Li, Tzu-Mao
Face registration deforms a template mesh to closely fit a 3D face scan, the quality of which commonly degrades in non-skin regions (e.g., hair, beard, accessories), because the optimized template-to-scan distance pulls the template mesh towards the noisy scan surface. Improving registration quality requires a clean separation of skin and non-skin regions on the scan mesh. Existing image-based (2D) or scan-based (3D) segmentation methods however perform poorly. Image-based segmentation outputs multi-view inconsistent masks, and they cannot account for scan inaccuracies or scan-image misalignment, while scan-based methods suffer from lower spatial resolution compared to images. In this work, we introduce a novel method that accurately separates skin from non-skin geometry on 3D human head scans. For this, our method extracts features from multi-view images using a frozen image foundation model and aggregates these features in 3D. These lifted 2D features are then fused with 3D geometric features extracted from the scan mesh, to then predict a segmentation mask directly on the scan mesh. We show that our segmentations improve the registration accuracy over pure 2D or 3D segmentation methods by 8.89% and 14.3%, respectively. Although trained only on synthetic data, our model generalizes well to real data.
Personalized Visual Dubbing through Virtual Dubber and Full Head Reenactment
(The Eurographics Association, 2025) Jeon, Bobae; Paquette, Eric; Mudur, Sudhir; Popa, Tiberiu; Ceylan, Duygu; Li, Tzu-Mao
Visual dubbing aims to modify facial expressions to ''lip-sync'' a new audio track. While person-generic talking head generation methods achieve expressive lip synchronization across arbitrary identities, they usually lack person-specific details and fail to generate high-quality results. Conversely, person-specific methods require extensive training. Our method combines the strengths of both methods by incorporating a virtual dubber, a person-generic talking head, as an intermediate representation. We then employ an autoencoder-based person-specific identity swapping network to transfer the actor identity, enabling fullhead reenactment that includes hair, face, ears, and neck. This eliminates artifacts while ensuring temporal consistency. Our quantitative and qualitative evaluation demonstrate that our method achieves a superior balance between lip-sync accuracy and realistic facial reenactment.
Real-time Neural Rendering of LiDAR Point Clouds
(The Eurographics Association, 2025) VANHERCK, Joni; Zoomers, Brent; Mertens, Tom; Jorissen, Lode; Michiels, Nick; Ceylan, Duygu; Li, Tzu-Mao
Static LiDAR scanners produce accurate, dense, colored point clouds, but often contain obtrusive artifacts which makes them ill-suited for direct display. We propose an efficient method to render more perceptually realistic images of such scans without any expensive preprocessing or training of a scene-specific model. A naive projection of the point cloud to the output view using 1×1 pixels is fast and retains the available detail, but also results in unintelligible renderings as background points leak between the foreground pixels. The key insight is that these projections can be transformed into a more realistic result using a deep convolutional model in the form of a U-Net, and a depth-based heuristic that prefilters the data. The U-Net also handles LiDAR-specific problems such as missing parts due to occlusion, color inconsistencies and varying point densities. We also describe a method to generate synthetic training data to deal with imperfectly-aligned ground truth images. Our method achieves real-time rendering rates using an off-the-shelf GPU and outperforms the state-of-the-art in both speed and quality.
TemPCC: Completing Temporal Occlusions in Large Dynamic Point Clouds captured by Multiple RGB-D Cameras
(The Eurographics Association, 2025) Mühlenbrock, Andre; Weller, Rene; Zachmann, Gabriel; Ceylan, Duygu; Li, Tzu-Mao
We present TemPCC, an approach to complete temporal occlusions in large dynamic point clouds. Our method manages a point set over time, integrates new observations into this set, and predicts the motion of occluded points based on the flow of surrounding visible ones. Unlike existing methods, our approach efficiently handles arbitrarily large point sets with linear complexity, does not reconstruct a canonical representation, and considers only local features. Our tests, performed on an Nvidia GeForce RTX 4090, demonstrate that our approach can complete a frame with 30,000 points in under 30 ms, while, in general, being able to handle point sets exceeding 1,000,000 points. This scalability enables the mitigation of temporal occlusions across entire scenes captured by multi-RGB-D camera setups. Our initial results demonstrate that self-occlusions are effectively completed and successfully generalized to unknown scenes despite limited training data.
Two-shot Shape and SVBRDF Reconstruction of Human Faces with Albedo-Conditioned Diffusion
(The Eurographics Association, 2025) Fan, Chongrui; Lin, Yiming; Lin, Arvin; Ghosh, Abhijeet; Ceylan, Duygu; Li, Tzu-Mao
Reconstructing 3D human heads with relightability has been a long-standing research problem. Most methods either require a complicated hardware setup for multiview capture or involve fitting a pre-learned morphable model, resulting in a loss of details. In our work, we present a two-step deep learning method that directly predicts the shape and SVBRDF of a subject's face given two images taken from each side of the face. We enhance SVBRDF prediction by first estimating the diffuse and specular albedo in image space, then generating texture maps in UV-space with a generative model. We also learn a 2D position map in UVspace for 3D geometry, eliminating the need for a morphable model. Contrary to single-image facial reconstruction methods, we obtain clear measurements on both sides of the face with two images. Our method outperforms state-of-the-art methods when rendering faces at extreme angles and provides texture maps that are directly usable in most rendering systems.
Neural Facial Deformation Transfer
(The Eurographics Association, 2025) Chandran, Prashanth; Ciccone, Loïc; Zoss, Gaspard; Bradley, Derek; Ceylan, Duygu; Li, Tzu-Mao
We address the practical problem of generating facial blendshapes and reference animations for a new 3D character in production environments where blendshape expressions and reference animations are readily available on a pre-defined template character. We propose Neural Facial Deformation Transfer (NFDT); a data-driven approach to transfer facial expressions from such a template character to new target characters given only the target's neutral shape. To accomplish this, we first present a simple data generation strategy to automatically create a large training dataset consisting of pairs of template and target character shapes in the same expression. We then leverage this dataset through a decoder-only transformer that transfers facial expressions from the template character to a target character in high fidelity. Through quantitative evaluations and a user study, we demonstrate that NFDT surpasses the previous state-of-the-art in facial expression transfer. NFDT provides good results across varying mesh topologies, generalizes to humanoid creatures, and can save time and cost in facial animation workflows.
3D Gabor Splatting: Reconstruction of High-frequnecy Surface Texture using Gabor Noise
(The Eurographics Association, 2025) Watanabe, Haato; Tojo, Kenji; Umetani, Nobuyuki; Ceylan, Duygu; Li, Tzu-Mao
3D Gaussian splatting has experienced explosive popularity in the past few years in the field of novel view synthesis. The lightweight and differentiable representation of the radiance field using the Gaussian enables rapid and high-quality reconstruction and fast rendering. However, reconstructing objects with high-frequency surface textures (e.g., fine stripes) requires many skinny Gaussian kernels because each Gaussian represents only one color if viewed from one direction. Thus, reconstructing the stripes pattern, for example, requires Gaussians for at least the number of stripes. We present 3D Gabor splatting, which augments the Gaussian kernel to represent spatially high-frequency signals using Gabor noise. The Gabor kernel is a combination of a Gaussian term and spatially fluctuating wave functions, making it suitable for representing spatial high-frequency texture. We demonstrate that our 3D Gabor splatting can reconstruct various high-frequency textures on the objects.

Browse

Browsing EG 2025 - Short Papers by Issue Date

Results Per Page

Sort Options