BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/Los_Angeles X-LIC-LOCATION:America/Los_Angeles BEGIN:DAYLIGHT TZOFFSETFROM:-0800 TZOFFSETTO:-0700 TZNAME:PDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0700 TZOFFSETTO:-0800 TZNAME:PST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260921T150539Z LOCATION:West Building\, Ballroom AB DTSTART;TZID=America/Los_Angeles:20250810T180000 DTEND;TZID=America/Los_Angeles:20250810T204500 UID:siggraph_SIGGRAPH 2025_sess284@linklings.com SUMMARY:Papers Fast Forward DESCRIPTION:Sponsored by Adobe Research\n\nAssetDropper: Asset Extraction via Diffusion Models with Reward-Driven Optimization\n\nAssetDropper is a novel framework for extracting standardized assets from reference images, addressing challenges such as occlusion and distortion. Leveraging both sy nthetic and real-world datasets, along with a reward-driven feedback mecha nism, it achieves state-of-the-art performance in asset extr...\n\n\nLanji ong Li (The Hong Kong University of Science and Technology (Guangzhou)); G uanhua Zhao (School of Electronic and Computer Engineering, Peking Univers ity); Lingting Zhu (The University of Hong Kong); Zeyu Cai (The Hong Kong University of Science and Technology (Guangzhou)); Lequan Yu (The Universi ty of Hong Kong); Jian Zhang (School of Electronic and Computer Engineerin g, Peking University); and Zeyu Wang (The Hong Kong University of Science and Technology (Guangzhou), The Hong Kong University of Science and Techno logy)\n---------------------\nMoVer: Motion Verification for Motion Graphi cs Animations\n\nLarge vision-language models often fail to capture spatio -temporal details in text-to-animation tasks. We introduce MoVer, a verifi cation system using first-order logic to check properties like timing and positioning in motion graphics animations. Integrated into an LLM pipeline , MoVer enables itera...\n\n\nJiaju Ma and Maneesh Agrawala (Stanford Univ ersity)\n---------------------\nEditDuet: A Multi-Agent System for Video N on-Linear Editing\n\nWe automate video nonlinear editing using a multi-age nt system. An Editor agent uses tools to create sequences from clips and i nstructions, while a Critic agent provides feedback in natural language. O ur learning-based approach enhances agent communication. Evaluations with an LLM-as-a-judge metric ...\n\n\nMarcelo Sandoval-Castañeda (TTIC, Adobe Research); Bryan Russell and Josef Sivic (Adobe Research); Gregory Shakhna rovich (TTIC); and Fabian David Caba Heilbron (Adobe Research)\n---------- -----------\nOrder Matters: Learning Element Ordering for Graphic Design G eneration\n\nWe propose a Generative Order Learner (GOL) that optimizes el ement ordering for graphic design generation. Our approach learns a conten t-aware neural order, which can significantly improve graphic generation q uality, generalize across different types of generative models and help de sign generators s...\n\n\nBo Yang and Ying Cao (ShanghaiTech University)\n ---------------------\nDesignManager: An Agent-Powered Copilot for Designe rs to Integrate AI Design Tools into Creative Workflows\n\nDesignManager i s an AI-powered design support system that functions as an interactive cop ilot throughout the creative workflow. With node-based visualization of de sign evolution and conversational interaction modes, it helps designers tr ack, modify, and branch their processes while providing context...\n\n\nWe itao You, Yinyu Lu, Zirui Ma, Nan Li, Mingxu Zhou, Xue Zhao, Pei Chen, and Lingyun Sun (Zhejiang University)\n---------------------\nHybrid Tours: A Clip-based System for Authoring Long-take Touring Shots\n\nWe propose Hyb rid Tours, a hybrid approach to creating long-take shots by combining shor t video clips in a virtual interface. We show that Hybrid Tours makes capt uring long-take touring shots much easier, and that clip-based authoring a nd reconstruction lead to higher-fidelity results at lower compu...\n\n\nX inrui Liu, Longxiulin Deng, and Abe Davis (Cornell University)\n---------- -----------\nMulti-Dimensional Procedural Wave Noise\n\nWe introduce a fas t, wave-based procedural noise model enabling precise spectral control in any dimension. Using precomputed wave functions and inverse Fourier transf orms, it supports Gaussian and non-Gaussian noises—including Gabor, Phasor , and novel recursive cellular patterns—making i...\n\n\nPascal Guehl (LIX - Ecole Polytechnique/CNRS, Institut Polytechnique de Paris); Rémi Allègr e (ICube, Université de Strasbourg; CNRS); Guillaume Gilet (Université de Sherbrooke); Basile Sauvage (ICube, Université de Strasbourg; CNRS); Marie -Paule Cani (LIX - Ecole Polytechnique/CNRS, Institut Polytechnique de Par is); and Jean-Michel Dischler (ICube, Université de Strasbourg; CNRS)\n--- ------------------\nNeurally Integrated Finite Elements for Differentiable Elasticity on Evolving Domains\n\nWe train a network to map signed distan ce fields to the quadrature points and weights of non-conforming numerical integration rule in a Mixed Finite Element formulation, enabling differen tiable elastic simulation over evolving domains. We demonstrate applicatio ns to image-guided material and topolog...\n\n\nGilles Daviet, Tianchang S hen, Nicholas Sharp, and David Levin (NVIDIA) and Gilles Daviet\n--------- ------------\nLightning-fast Boundary Element Method\n\nWe introduce an in verse-LU preconditioner to solve for the typical asymmetric and dense matr ices generated by boundary element methods (BEM). The computational effici ency and low memory requirements of our approach conspire to scale up to m illions of degrees of freedom, with orders of magnitude spee...\n\n\nJiong Chen (INRIA Saclay); Florian Schäfer (Georgia Institute of Technology); a nd Mathieu DESBRUN (INRIA Saclay, Ecole Polytechnique)\n------------------ ---\nMonte Carlo PDE simulation in participating media\n\nWe solve partial differential equations in domains involving complex microparticle geometr y that is impractical, or intractable, to model explicitly. Drawing inspir ation from volume rendering, we treat the domain as a participating medium with stochastic microparticle geometry and develop a volumetr...\n\n\nBai ley Miller (Carnegie Mellon University), Rohan Sawhney (NVIDIA), and Keena n Crane and Ioannis Gkioulekas (Carnegie Mellon University)\n------------- --------\nConformal First Passage for Epsilon-free Walk-on-Spheres\n\nWe p resent a novel Monte Carlo approach to solve boundary integral equations w ith Dirichlet boundary conditions in two dimensions. While Walk-on-Spheres uses largest empty circles, which touch the boundary in only one point, w e utilize semicircles and circle sectors that share one or two boundary ed ...\n\n\nPaul Himmler and Tobias Günther (Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU))\n---------------------\nGuiding-Based Importance Sampling for Walk on Stars\n\nWalk on stars (WoSt) has shown its power in being applied to Monte Carlo methods for solving PDEs but the sampling tec hniques in WoSt are not satisfactory, leading to high variance. Inspired b y Monte Carlo rendering, we propose a guiding-based importance sampling me thod to reduce the variance of WoS...\n\n\nTianyu Huang and Jingwang Ling (School of Software and BNRist, Tsinghua University); Shuang Zhao (Univers ity of California Irvine); and Feng Xu (School of Software and BNRist, Tsi nghua University)\n---------------------\nMotion Inversion for Video Custo mization\n\nWe propose Motion Embeddings for video generation, enabling pr ecise motion in video transfer across diverse scenes and objects. These em beddings disentangle motion from appearance, preserving original dynamics while adapting to new prompts. Experiments show that our method achieved h igh-quality, pro...\n\n\nLuozhou Wang, Ziyang Mai, and Guibao Shen (Hong K ong University of Science and Technology, Guangzhou); Yixun Liang (Hong Ko ng University of Science and Technology); Xin Tao, Pengfei Wan, and Di Zha ng (Kuaishou Technology); Yijun Li (Adobe Research); and Yingcong Chen (Ho ng Kong University of Science and Technology, Guangzhou)\n---------------- -----\nGaVS: 3D-Grounded Video Stabilization via Temporally-Consistent Loc al Reconstruction and Rendering\n\nGaVS: Transform unstable shaky videos i nto smooth, professional-quality footage. We design novel 3D rednering tec hnology that preserves the motion intent while eliminating shakes and dist ortions—no cropping, no distortion and workable under dynamics and intense motions. GaVS delivers natural-l...\n\n\nZinuo You (ETH Zürich, Huawei Re search Zürich); Stamatios Georgoulis (Huawei Research Zürich); Anpei Chen (ETH Zürich, University of Tuebingen); Siyu Tang (ETH Zürich); and Dengxin Dai (Huawei Research Zürich)\n---------------------\nNested Attention: Se mantic-aware Attention Values for Concept Personalization\n\nThis paper in troduces Nested Attention, a mechanism that improves text-to-image persona lization by injecting query-dependent subject features into cross-attentio n layers, achieving strong identity preservation and prompt alignment. The method maintains the model’s prior, enabling multi-subject...\n\n\nOr Pat ashnik (Tel Aviv University, Snap); Rinon Gal (Tel Aviv University); Danii l Ostashev, Sergey Tulyakov, and Kfir Aberman (Snap); and Daniel Cohen-Or (Tel Aviv University, Snap)\n---------------------\nTokenVerse: Versatile Multi-concept Personalization in Token Modulation Space\n\nTokenVerse extr acts complex visual elements from images by identifying semantic direction s in per-token modulation space of DiT models for each word in the image c aption. It's capable of combining concepts from multiple sources by adding corresponding directions, enabling flexible generation of new ...\n\n\nDa niel Garibi (Tel Aviv University, DeepMind); Shahar Yadin (Technion - Isra el Institute of Technology, DeepMind); Roni Paiss, Omer Tov, Shiran Zada, and Ariel Ephrat (DeepMind); Tomer Michaeli (Technion - Israel Institute o f Technology, DeepMind); Inbar Mosseri (DeepMind); and Tali Dekel (Weizman n Institute of Science, DeepMind)\n---------------------\nDynamic Concepts Personalization from Single Videos\n\nPersonalizing text-to-video models is challenging because dynamic concepts require capturing both appearance and motion. We propose Set-and-Sequence, a framework that personalizes DiT -based video models by first learning an identity LoRA basis from unordere d frames, then fine-tuning coefficients wit...\n\n\nRameen Abdal, Or Patas hnik, Ivan Skorokhodov, Willi Menapace, Aliaksandr Siarohin, Sergey Tulyak ov, Daniel Cohen-Or, and Kfir Aberman (Snap)\n---------------------\nFlexi Act: Towards Flexible Action Control in Heterogeneous Scenarios\n\nWe prop ose FlexiAct, an image animation framework that transfers actions from a r eference video to any target image, enabling variations in layout, viewpoi nt, and skeletal structure while maintaining identity consistency.\n\n\nSh iyi Zhang (Tsinghua University); Junhao Zhuang, Zhaoyang Zhang, and Ying S han (Tencent); and Yansong Tang (Tsinghua University)\n------------------- --\nAdaptive Phase-Field-FLIP for Very Large Scale Two-Phase Fluid Simulat ion\n\nWe present an algorithm for simulating large-scale, violently turbu lent two-phase flows—such as breaking ocean waves, tsunamis, and asteroid impacts—at extreme resolutions of the coupled water-air velocity field. Th is is achieved by integrating a new multiphase FLIP variant with highly e. ..\n\n\nBernhard Braun (Technical University Munich), Jan Bender (RWTH Aac hen University), and Nils Thuerey (Technical University Munich)\n--------- ------------\nFast Subspace Fluid Simulation with a Temporally-Aware Basis \n\nWe introduce a novel reduced-order fluid simulation technique leveragi ng Dynamic Mode Decomposition (DMD) to enable fast, memory-efficient, and user-controllable subspace simulation. Combining spatial ROM compression w ith spectral physical insights, our method excels in animation, real-time interact...\n\n\nSiyuan Chen (University of Toronto, Shanghai Jiao Tong Un iversity) and Yixin Chen, Jonathan Panuelos, Otman Benchekroun, Yue Chang, Eitan Grinspun, and Zhecheng Wang (University of Toronto)\n-------------- -------\nGaussian Fluids: A Grid-Free Fluid Solver based on Gaussian Spati al Representation\n\nWe present a grid-free fluid simulator featuring a no vel Gaussian spatial representation (GSR) for velocity field. The advantag es of GSR over traditional Lagrangian/Eulerian data structures are 4-folde d: memory compactness, spatial adaptivity, vorticity preservation and cont inuous differentiability....\n\n\nJingrui Xing (School of Intelligence Sci ence and Technology, Peking University); Bin Wang (Independent); and Mengy u Chu and Baoquan Chen (Peking University, State Key Laboratory of General Artificial Intelligence)\n---------------------\nQuadtree Tall Cells for Eulerian Liquid Simulation\n\nThis paper introduces a novel grid structure that extends tall cell methods for efficient deep water simulation. Unlik e previous methods, our approach subdivides tall cells horizontally, allow ing for more aggressive adaptivity. We demonstrate that this novel form of adaptivity delivers superior perf...\n\n\nFumiya Narita (The University o f Tokyo, GAME FREAK Inc.); Nimiko Ochiai (GAME FREAK Inc.); Takashi Kanai (The University of Tokyo); and Ryoichi Ando (Unaffiliated)\n-------------- -------\nA Neural Particle Level Set Method for Dynamic Interface Tracking \n\nWe propose Neural PLS, a neural particle level-set method for tracking and evolving dynamic neural representations. Oriented particles serve as interface trackers and sampling seeders, enabling efficient evolution on a multi-resolution grid-hash structure. Our approach integrates traditional PLS and...\n\n\nDuowen Chen (Georgia Institute of Technology), Junwei Zho u (University of Michigan), Bo Zhu (Georgia Institute of Technology), and Duowen Chen\n---------------------\nMotion Control via Metric-Aligning Mot ion Matching\n\nMetric-Aligning Motion Matching (MAMM) is a novel method f or controlling motion sequences using sketches, labels, audio, or another motion sequence without requiring training or annotations. By aligning wit hin-domain distances, MAMM provides a flexible and efficient solution for motion control acros...\n\n\nNaoki Agata and Takeo Igarashi (The Universit y of Tokyo)\n---------------------\nAutoKeyframe: Autoregressive Keyframe Generation for Human Motion Synthesis and Editing\n\nWe present AutoKeyfra me, a novel framework that simultaneously accepts dense and sparse control signals for motion generation by generating keyframes directly. Our metho d reduces manual efforts for keyframing while maintaining precise controll ability, using an autoregressive diffusion model and a ne...\n\n\nBowen Zh eng and Ke Chen (Zhejiang University; State Key Laboratory of CAD&CG, Zhej iang University); Yuxin Yao (University of Cambridge, Department of Engine ering); Zijiao Zeng and Xinwei Jiang (Tencent Games Digital Content Techno logy Center); He Wang (UCL Centre for Artificial Intelligence, Department of Computer Science, University College London); Joan Lasenby (University of Cambridge, Department of Engineering); and Xiaogang Jin (Zhejiang Unive rsity; State Key Laboratory of CAD&CG, Zhejiang University)\n------------- --------\nKinematic Motion Retargeting for Contact-Rich Anthropomorphic Ma nipulations\n\nWe present a simple, but effective framework for kinematica lly retargeting contact-rich anthropomorphic hand-object manipulations by exploiting contact areas. We reliably retarget contact area data between d iverse hands using a novel non-isometric shape matching process and genera te high quality res...\n\n\nArjun Lakshmipathy, Jessica Hodgins, and Nancy Pollard (Carnegie Mellon University) and Arjun Lakshmipathy\n------------ ---------\nAnyTop: Character Animation Diffusion with Any Topology\n\nAnyT op generates motion for diverse character skeletons using only skeletal st ructure as input. This diffusion model incorporates topology information a nd textual joint descriptions to learn semantic correspondences across dif ferent skeletons. It generalizes with minimal training examples and suppor ...\n\n\nInbar Gat, Sigal Raab, Guy Tevet, Yuval Reshef, Amit Haim Bermano , and Daniel Cohen-Or (Tel Aviv University)\n---------------------\nAdvanc ing GPU IPC for Stiff Affine-Deformable Simulation\n\nWe present a GPU-opt imized IPC framework achieving up to 10× speedup across soft, stiff, and h ybrid simulations. Key innovations include a connectivity-enhanced MAS pre conditioner, a parallel-friendly inexact strain limiting energy, and a has h-based two-level reduction strategy for fast Hes-\nsian as...\n\n\nKemeng Huang (Carnegie Mellon University, The University of Hong Kong); Xinyu Lu (TransGP); Huancheng Lin (Carnegie Mellon University, The University of H ong Kong); Taku Komura (The University of Hong Kong); Minchen Li (Carnegie Mellon University); and Kemeng Huang\n---------------------\nJGS2: Near S econd-order Converging Jacobi/Gauss-Seidel for GPU Elastodynamics\n\nThis paper presents a new GPU simulation algorithm, which converges as fast as global Newton's method and as efficient as Jacobi method.\n\n\nLei Lan (Un iversity of Utah; State Key Lab of CAD&CG, Zhejiang University); Zixuan Lu and Chun Yuan (University of Utah); Weiwei Xu (State Key Lab of CAD&CG, Z hejiang University, China); Hao Su (UCSD); Huamin Wang (Style3D Research); Chenfanfu Jiang (UCLA); and Yin Yang (University of Utah)\n-------------- -------\nHyper-Dimensional Deformation Simulation\n\nEver feel like three dimensions isn't quite enough? We performed the analysis necessary to simu late the motion of deformables in four spatial dimensions! Along the way, we developed techniques for generating simulation-ready hyper-meshes, anal yzing hyper-dimensional deformation energies, and detecti...\n\n\nAlvin Sh i, Haomiao Wu, and Theodore Kim (Yale University)\n---------------------\n Transformer IMU Calibrator: Dynamic On-body IMU Calibration for Inertial M otion Capture\n\nThis work proposes a dynamic calibration system for inert ial motion capture, which can dynamically remove non-static IMU drift and sensor-body offset during usage, enable user-friendly calibration (without T-pose and IMU heading reset), and ensure long-term robustness.\n\n\nChen gxu Zuo, Jiawei Huang, Xiao Jiang, and Yuan Yao (Xiamen University); Xiang ren Shi (Bournemouth University); Rui Cao (Xiamen University); Xinyu Yi an d Feng Xu (Tsinghua University); Shihui Guo (Xiamen University); and Yipen g Qin (Cardiff University)\n---------------------\nA Platform for Interact ive AI Character Experiences\n\nWe present a platform for creating believa ble, conversational digital characters that combine conversational AI, spe ech, animation, memory, personality, and emotions. Demonstrated through Di gital Einstein, our system enables interactive, story-driven experiences a nd generalizes to any character, mak...\n\n\nRafael Wampfler, Chen Yang, D illon Elste, Nikola Kovacevic, Philine Witzig, and Markus Gross (ETH Züric h)\n---------------------\nLearning to Draw Is Learning to See: Analyzing Eye Tracking Patterns for Assisted Observational Drawing\n\nWe present an image-to-image drawing setup capturing eye tracking and stroke data across 156 drawings from 10 artists. Our findings reveal consistent fixation pat terns, strong gaze–stroke correlations, and structured drawing sequences, offering new insights into professional observation strate...\n\n\nFengqi LIU, Longji Huang, and Zhengyu Huang (The Hong Kong University of Science and Technology (Guangzhou)) and Zeyu Wang (The Hong Kong University of Sci ence and Technology (Guangzhou), The Hong Kong University of Science and T echnology)\n---------------------\nHOIGaze: Gaze Estimation During Hand-Ob ject Interactions in Extended Reality Exploiting Eye-Hand-Head Coordinatio n\n\nWe present HOIGaze – a novel approach for gaze estimation during hand -object interactions in extended reality. HOIGaze features: 1) a novel hie rarchical framework that first recognises attended hand and then estimates gaze; 2) a new gaze estimator combining CNN, GCN, and cross-modal Transfo rme...\n\n\nZhiming Hu (University of Stuttgart, The Hong Kong University of Science and Technology (Guangzhou)); Daniel Haeufle (University of Tueb ingen, The Center for Bionic Intelligence Tuebingen Stuttgart); Syn Schmit t (University of Stuttgart, The Center for Bionic Intelligence Tuebingen S tuttgart); and Andreas Bulling (University of Stuttgart)\n---------------- -----\nForceGrip: Reference-Free Curriculum Learning for Realistic Grip Fo rce Control in VR Hand Manipulation\n\nForceGrip is a reference-free reinf orcement learning-based agent for realistic VR hand manipulation. It uses a progressive curriculum (finger positioning, intention adaptation, dynami c stabilization) and physics simulation to convert VR controller inputs in to faithful grip forces. In user studies, p...\n\n\nDongHeun Han (Kyung He e University), Byungmin Kim (Korea University), RoUn Lee (Kyung Hee Univer sity), KyeongMin Kim (Korea University), Hyoseok Hwang (Kyung Hee Universi ty), and HyeongYeop Kang (Korea University)\n---------------------\nVirCHE W Reality: On-Face Kinesthetic Feedback for Enhancing Food-Intake Experien ce in Virtual Reality\n\nThis paper presents VirCHEW Reality, a face-worn haptic device for virtual food intake in VR. It uses pneumatic actuation t o simulate food textures, enhancing the chewing experience. User studies d emonstrated its effectiveness in providing distinct kinesthetic feedback a nd improving virtual eating e...\n\n\nQingqin Liu, Ziqi Fang, and Jiayi Wu (School of Creative Media, City University of Hong Kong); Shaoyu Cai (Nat ional University of Singapore); Jianhui Yan (School of Creative Media, Cit y University of Hong Kong); Tiande Mo and Shuk Ching CHAN (Hong Kong Produ ctivity Council); and Kening Zhu (City University of Hong Kong)\n--------- ------------\nSplat and Replace: 3D Reconstruction with Repetitive Element s\n\nWe leverage repetitions in 3D scenes to improve reconstruction in low -quality parts due to poor coverage and occlusions. Our methods segments t he repetitions, registers them together, and optimizes a shared representa tion with multi-view information flowing from all repetitions, improving t he recons...\n\n\nNicolas Violante, Andréas Meuleman, and Alban Gauthier ( INRIA, Université Côte d'Azur); Fredo Durand (Massachusetts Institute of T echnology (MIT)); Thibault Groueix (Adobe Research); and George Drettakis (INRIA, Université Côte d'Azur)\n---------------------\nSpotLessSplats: Ig noring Distractors in 3D Gaussian Splatting\n\n3D Gaussian Splatting (3DGS ) enables fast 3D reconstruction and rendering but struggles with real-wor ld captures due to transient elements and lighting changes. We introduce S potLessSplats, which leverages semantic features from foundation models an d robust optimization to remove transient effects, ...\n\n\nSara Sabour (G oogle Inc - Deepmind, University of Toronto); Lily Goli (University of Tor onto); George Kopanas (Runway); Mark Matthews and Dmitry Lagun (Google Inc - Deepmind); Leonidas Guibas (Google Inc - Deepmind, Stanford University) ; Alec Jacobson (University of Toronto); David Fleet (Google Inc - Deepmin d, University of Toronto); Andrea Tagliasacchi (Google Inc - Deepmind, Sim on Fraser University); and Sara Sabour and Lily Goli\n-------------------- -\nRadiance Surfaces: Optimizing Surface Representations with a 5D Radianc e Field Loss\n\nWe present a simple and fast method to reconstruct radianc e surfaces by directly supervising the radiance field via image projection . \nUnlike volumetric approaches, we move alpha blending and ray marching from image formation into loss computation. \nThis simple modification ena bles high-quality surf...\n\n\nZiyi Zhang and Nicolas Roussel (EPFL); Thom as Muller, Tizian Zeltner, Merlin Nimier-David, and Fabrice Rousselle (NVI DIA); and Wenzel Jakob (EPFL)\n---------------------\nEchoes of the Colise um: Towards 3D Live streaming of Sports Events\n\nWe present a revolutiona ry method for experiencing live sports in stunning 3D, redefining the way games are seen, through immersive, interactive replays. Alongside, we rele ase a large-scale synthetic dataset built to benchmark realism, motion, an d human interaction in dynamic scenes, to fuel the nex...\n\n\nJunkai Huan g, Saswat Subhajyoti Mallick, Alejandro Amat, Marc Ruiz Olle, Albert Mosel la-Montoro, Bernhard Kerbl, Francisco Vicente Carrasco, and Fernando De la Torre (Carnegie Mellon University)\n---------------------\nNeST: Neural S tress Tensor Tomography by leveraging 3D Photoelasticity\n\nNeST enables n on-destructive 3D stress analysis of transparent objects using the polariz ation of light. Traditional 2D methods require destructively slicing the o bject. Instead, we reconstruct the entire 3D stress field by jointly handl ing phase unwrapping and tensor tomography using neural implicit...\n\n\nA kshat Dave (Massachusetts Institute of Technology Media Lab), Tianyi Zhang (Rice University), Aaron Young and Ramesh Raskar (Massachusetts Institute of Technology Media Lab), Wolfgang Heidrich (King Abdullah University of Science and Technology), Ashok Veeraraghavan (Rice University), and Akshat Dave\n---------------------\nRevisiting Tradition and Beyond: A Customize d Bilateral Filtering Framework for Point Cloud Denoising\n\nTo combine de ep learning's generalization with traditional methods' interpretability, w e propose CustomBF—a hybrid framework that customizes bilateral filter com ponents per point. By addressing key limitations of the classic bilateral filter, CustomBF achieves robust, interpretable, and effect...\n\n\nPeng L i, Zeyong Wei, Honghua Chen, Xuefeng Yan, and Mingqiang Wei (Nanjing Unive rsity of Aeronautics and Astronautics)\n---------------------\nA Closest P oint Method for PDEs on Manifolds with Interior Boundary Conditions for Ge ometry Processing\n\nGeometry processing often requires the solution of PD Es with boundary conditions on the manifold’s interior. However, input man ifolds can take many forms, each requiring specialized discretizations. In stead, we develop a unified framework for general manifold representations by extending the c...\n\n\nNathan King (University of Waterloo), Haozhe S u (LightSpeed Studios), Mridul Aanjaneya (Rutgers University), Steven Ruut h (Simon Fraser University), Christopher Batty (University of Waterloo), a nd Nathan King\n---------------------\nA Divide-and-Conquer Approach for G lobal Orientation of Non-Watertight Scene-Level Point Clouds with 0-1 Inte ger Optimization\n\nWe propose a divide-and-conquer approach for orienting large-scale, non-watertight point clouds. The scene is first segmented in to blocks, and normal orientations are estimated independently within each block. These local orientations are then globally unified through a graph -based formulation, solv...\n\n\nZhuodong Li, Fei Hou, and Wencheng Wang ( Institute of Software, Chinese Academy of Sciences; University of Chinese Academy of Sciences); Xuquan Lu (The University of Western Australia); and Ying He (Nanyang Technological University)\n---------------------\nFarada y Cage Estimation of Normals for Point Clouds and Ribbon Sketches\n\nWe pr opose a novel method for normal estimation of unoriented point clouds and VR ribbon sketches that leverages a modeling of the Faraday cage effect. O ur method is uniquely robust to the presence of interior structures and ar tifacts, producing superior surfacing output when combined with Poisson S. ..\n\n\nDaniel Scrivener, Daniel Cui, and Ellis Coldren (Boston University ); Mazdak Abulnaga (Massachusetts Institute of Technology (MIT), Harvard M edical School); Mikhail Bessmeltsev (Universite de Montreal); and Edward C hien (Boston University)\n---------------------\nA Polyhedral Construction of Empty Spheres in Discrete Distance Fields\n\nSpheres that are disjoint from a given union of spheres can be computing by solving a convex hull p roblem. This can be exploited for contouring discretely sampled signed dis tance functions.\n\n\nMax Kohlbrenner and Marc Alexa (Technical University of Berlin)\n---------------------\nVariable Shared Template for Consisten t Non-rigid ICP\n\nWe propose a novel ICP framework that jointly optimizes a shared template and instance-wise deformations. Our approach automatica lly captures common shape features from input shapes, achieving state-of-t he-art accuracy and consistency while eliminating the need to carefully se lect a preset template ...\n\n\nYucheol Jung, Hyomin Kim, Hyejeong Yoon, Y oonha Hwang, and Seungyong Lee (POSTECH)\n---------------------\nPhysics-i nspired Estimation of Optimal Cloth Mesh Resolution\n\nWe propose a method to estimate optimal cloth mesh resolution based on material stiffness and boundary conditions like shirring or stitching, and dynamic wrinkles from motion-induced collisions. To ensure smooth resolution transitions, we ca lculate transition distances and generate a mesh sizing map...\n\n\nDiyang Zhang, Zhendong Wang, and Zegao Liu (Style3D Research); Xinming Pei (Stat e Key Laboratory of CAD & CG, Zhejiang University; Style3D Research); Weiw ei Xu (State Key Laboratory of CAD & CG, Zhejiang University); and Huamin Wang (Style3D Research)\n---------------------\nAutomated Task Scheduling for Cloth and Deformable Body Simulations in Heterogeneous Computing Envir onments\n\nThis paper introduces an automated scheduling framework to opti mize cloth and deformable simulations across heterogeneous computing devic es. Using an enhanced HEFT algorithm and asynchronous iteration methods, o ur approach minimizes communication delays and maximizes parallelism. our experiments dem...\n\n\nChengzhu He (Xiamen University, Style3D Research); Zhendong Wang (Style3D Research); Zhaorui Meng, Junfeng Yao, and Shihui G uo (Xiamen University); and Huamin Wang (Style3D Research)\n-------------- -------\nHigh-performance CPU Cloth Simulation Using Domain-decomposed Pro jective Dynamics\n\nThis paper presents a CPU-based cloth simulation algor ithm that partitions garment models into domains that can be processed by each individual CPU core. Using projective dynamics with domain-level para llelization, this method achieves high performance comparable to GPU meth ods and runs about an or...\n\n\nZixuan Lu, Ziheng Liu, and Lei Lan (Unive rsity of Utah); Huamin Wang (Style3D Research); Yuko Ishiwaka (SoftBank); Chenfanfu Jiang (UCLA); Kui Wu (LightSpeed Studios); and Yin Yang (Univers ity of Utah)\n---------------------\nReal-Time Knit Deformation and Render ing\n\nIn this work, we introduce the first real-time framework that integ rates yarn-level simulation with fiber-level rendering. The whole system p rovides real-time performance and has been evaluated through various appli cation scenarios, including knit simulation for small patches and full gar ments and y...\n\n\nTao Huang (LightSpeed Studios), Haoyang Shi (Universit y of Utah), Mengdi Wang and Yuxing Qiu (LightSpeed Studios), Yin Yang (Uni versity of Utah), and Kui Wu (LightSpeed Studios)\n---------------------\n Progressive Dynamics++: A Framework for Stable, Continuous, and Consistent Animation Across Resolution and Time\n\nWe propose a general framework, P rogressive Dynamics++, for constructing a family of progressive dynamics i ntegration methods that advance physical simulation states forward in both time and spatial resolution. We analyze requirements for stable, continuo us, and consistent level-of-detail animation ...\n\n\nJiayi Eris Zhang (St anford University, Adobe); Doug James (Stanford University); and Danny Kau fman (Adobe)\n---------------------\nLifting the Winding Number: Precise D iscontinuities in Neural Fields for Physics Simulation\n\nWe designed a ne ural field capable of capturing a diverse family of discontinuities, enabl ing the simulation of cuts in thin-walled deformable structures. By liftin g input coordinates using generalized winding numbers, our approach models discontinuities explicitly and flexibly, supporting accurate,...\n\n\nYue Chang, Mengfei Liu, and Zhecheng Wang (University of Toronto); Peter Yich en Chen (MIT CSAIL); and Eitan Grinspun (University of Toronto)\n--------- ------------\nLight Pipe Holographic Display: Bandwidth-preserved Kaleidos copic Guiding for AR Glasses\n\nWe present and analyze a holographic augme nted reality display with the bandwidth-preserved guiding method using a l ight pipe. We propose the use of light pipe to spatially relocate the ligh t engine from the image combiner at the front-module, enabling enhanced we ight distribution and obstruction-fr...\n\n\nMinseok Chae, Chun Chen, Seun g-Woo Nam, and Yoonchan Jeong (Seoul National University)\n--------------- ------\nHoloChrome: Polychromatic Illumination for Speckle Reduction in Ho lographic Near-Eye Displays\n\nHoloChrome introduces a novel holographic d isplay method by multiplexing multiple wavelengths and two spatial light m odulators to enhance image quality. By moving beyond standard three-color primary systems, it significantly reduces speckle noise while preserving n atural depth cues while achieving m...\n\n\nFlorian Schiffers (Amazon Prim e Video, Northwestern University); Grace Kuo, Nathan Matsuda, Douglas Lanm an, and Oliver Cossairt (Meta Reality Labs); and Florian Schiffers and Oli ver Cossairt\n---------------------\nNoise-Coded Illumination for Forensic and Photometric Video Analysis\n\nVideo forensics, which focuses on ident ifying fake or manipulated video, is becoming increasingly difficult with the development of more advanced video editing techniques. We show how cod ing near-imperceptible, noise-like modulations into the illumination of a scene can create information asymmetry ...\n\n\nPeter Michael (Cornell Uni versity), Zekun Hao (Cornell Tech), Serge Belongie (University of Copenhag en), Abe Davis (Cornell University), and Peter Michael\n------------------ ---\nCollaborative On-Sensor Array Cameras\n\nWe introduce a collaborative metalens array comprising over 100-million nanopillars for broadband imag ing. The proposed array camera is only a few millimeters flat and employs a non-generative reconstruction method, which performs favorably and witho ut hallucinations, irrespective of the scene illum...\n\n\nJipeng Sun (Pri nceton University), Kaixuan Wei (King Abdullah University of Science and T echnology (KAUST)), Thomas Eboli (Université Paris-Saclay), Congli Wang an d Cheng Zheng (Princeton University), Zhihao Zhou and Arka Majumdar (Unive rsity of Washington), Wolfgang Heidrich (King Abdullah University of Scien ce and Technology (KAUST)), and Felix Heide (Princeton University)\n------ ---------------\nA Monte Carlo Rendering Framework for Simulating Optical Heterodyne Detection\n\nWe present a general spectral-domain simulation fr amework for optical heterodyne detection (OHD), extending path integral re ndering to capture power spectral density of OHD. Unlike existing domain-s pecific tools, our approach supports diverse scenes and applications. We v alidate it against real-worl...\n\n\nJuhyeon Kim (Dartmouth College); Crai g Benko, Magnus Wrenninge, Ryusuke Villemin, and Zeb Barber (Aurora Innova tion); and Wojciech Jarosz and Adithya Pediredla (Dartmouth College)\n---- -----------------\nGaussian Wave Splatting for Computer-Generated Holograp hy\n\nWe develop novel and efficient computer-generated holography algorit hms, dubbed Gaussian Wave Splatting, that transform Gaussian-based scene r epresentations into holograms. We derive a closed-form 2D Gaussian-to-holo gram transform supporting occlusions and alpha blending, along with an eff icient, ea...\n\n\nSuyeon Choi, Brian Chao, Jacqueline Yang, Manu Gopakuma r, and Gordon Wetzstein (Stanford University)\n---------------------\nFaci al Appearance Capture at Home with Patch-Level Reflectance Prior\n\nGiven a single co-located smartphone video captured in a dim room as the input, our method can reconstruct high-quality facial assets within the distribut ion modeled by a diffusion prior trained on Light Stage scans, which can b e exported to common graphics engines like Blender for photo-realistic r.. .\n\n\nYuxuan Han and Junfeng Lyu (Tsinghua University); Kuan Sheng (Shang haiTech University; Deemos Technology Co., Ltd.); Minghao Que (Tsinghua Un iversity); Qixuan Zhang (ShanghaiTech University; Deemos Technology Co., L td.); Lan Xu (ShanghaiTech University); and Feng Xu (Tsinghua University)\ n---------------------\nText-based Animatable 3D Avatars with Morphable Mo del Alignment\n\nAnimPortrait3D is a novel method for text-based, realisti c, animatable 3DGS avatar generation with morphable model alignment. To ad dress ambiguities in diffusion predictions during 3D distillation, we intr oduce key strategies: initializing a 3D avatar with robust appearance and geometry, and leverag...\n\n\nYiqian Wu (ETH Zürich; State Key Lab of CAD and CG, Zhejiang University); Malte Prinzler (ETH Zürich); Xiaogang Jin (S tate Key Lab of CAD and CG, Zhejiang University); and Siyu Tang (ETH Züric h)\n---------------------\nLAM: Large Avatar Model for One-shot Animatable Gaussian Head\n\nLAM is an innovative Large Avatar Model for animatable G aussian head reconstruction from a single image in seconds. Our Gaussian h eads are immediately animatable and renderable without additional networks or post-processing. This allows seamless integration into existing render ing pipelines, ensurin...\n\n\nYisheng He, Xiaodong Gu, Xiaodan Ye, Chao X u, Zhengyi Zhao, Yuan Dong, Weihao Yuan, Zilong Dong, and Liefeng Bo (Alib aba Group)\n---------------------\nSOAP: Style-Omniscient Animatable Portr aits\n\nSOAP awakens the 3D princess from 2D stylized photos. Unlike other works that directly drive the 2D photos, SOAP reconstructs well-rigged 3D avatars, with detailed geometry and all-around texture, from just a singl e stylized picture.\n\n\nTingting Liao and Yujian Zheng (Mohamed bin Zayed University of Artificial Intelligence); Yuliang Xiu (Westlake University) ; Adilbek Karmanov (Mohamed bin Zayed University of Artificial Intelligenc e); Liwen Hu (Pinscreen); Leyang Jin (Mohamed bin Zayed University of Arti ficial Intelligence); and Hao Li (Mohamed bin Zayed University of Artifici al Intelligence, Pinscreen)\n---------------------\n3DGH: 3D Head Generati on with Composable Hair and Face\n\nWe present 3DGH, a generative model th at creates realistic 3D human heads with composable hair and face componen ts. By modeling both the separation and correlation between hair and face in a generative paradigm, it enables high-quality, full-head synthesis and flexible 3D hairstyle editing with stro...\n\n\nChengan He (Yale Universi ty); Junxuan Li (Meta Codec Avatars Lab); Tobias Kirschstein and Artem Sev astopolskiy (Technical University of Munich); Shunsuke Saito, Qingyang Tan , Javier Romero, and Chen Cao (Meta Codec Avatars Lab); Holly Rushmeier (Y ale University); and Giljoo Nam (Meta Codec Avatars Lab)\n---------------- -----\nFacial Microscopic Structures Synthesis from a Single Unconstrained Image\n\nOur framework can efficiently synthesize facial microstructure f rom an unconstrained facial image via differentiable optimization. We prop ose neural wrinkle simulation for differentiable microstructure parameteri zation, and direction distribution similarity to align features with blurr y image patche...\n\n\nYouyang Du and Lu Wang (Shandong University) and Be ibei Wang (Nanjing University)\n---------------------\nNAM: Neural Adjoint Maps for refinement of shape correspondences\n\nWe introduce Neural Adjoi nt Maps, a novel representation for correspondences between 3D shapes. Bui lt on and extending the functional map framework, our approach enables acc urate, non-linear refinement of shape matching across meshes and point clo uds, setting a new standard in diverse scenarios and ...\n\n\nGiulio Vigan ò (Università di Milano Bicocca), Maks Ovsjanikov (Centre National de la R echerche Scientifique - Laboratoire d'informatique de l'École Polytechniqu e (LIX)), and Simone Melzi (Università di Milano Bicocca)\n--------------- ------\nGenAnalysis: Joint Shape Analysis by Learning Man-Made Shape Gener ators with Deformation Regularizations\n\nWe present GenAnalysis, an impli cit shape generation framework enabling joint shape matching and consisten t segmentation by enforcing as-affine-as-possible (AAAP) deformations via regularization loss in latent space. It enables shape analysis via extract ing and analysing shape variations in the tang...\n\n\nYuezhi Yang and Hai tao Yang (University of Texas at Austin), George Kiyohiro Nakayama (Stanfo rd University), Xiangru Huang (Westlake University), Leonidas Guibas (Stan ford University), and Qixing Huang (University of Texas at Austin)\n------ ---------------\nMASH: Masked Anchored SpHerical Distances for 3D Shape Re presentation and Generation\n\nWe introduce Masked Anchored SpHerical Dist ances (MASH), a novel multi-view and parametrized representation of 3D sha pes. MASH is versatilefor multiple applications including surface reconstr uction, shape generation, completion, and blending, achieving superior per formance thanks to its unique repre...\n\n\nChanghao Li and Yu Xin (Univer sity of Science and Technology of China); Xiaowei Zhou (State Key Laborato ry of CAD & CG, Zhejiang University); Ariel Shamir (Reichman University); Hao Zhang (Simon Fraser University); Ligang Liu (University of Science and Technology of China); and Ruizhen Hu (Shenzhen University)\n------------- --------\nUnsupervised Decomposition of 3D Shapes into Expressive and Edit able Extruded Profile Primitives\n\n3D2EP transforms 3D shapes into expres sive, editable primitives by extruding 2D profiles along 3D curves. This a pproach creates compact, interpretable representations that support intuit ive editing and flexible redesign. It delivers high fidelity and efficienc y, outperforming existing methods across...\n\n\nChunyi Sun (Australian Na tional University); Junlin Han and Runjia Li (University of Oxford); and W eijian Deng, Dylan Campbell, and Stephen Gould (Australian National Univer sity)\n---------------------\nOctGPT: Octree-based Multiscale Autoregressi ve Models for 3D Shape Generation\n\nOctGPT is a novel multiscale autoregr essive model for 3D shape generation. It introduces hierarchical serialize d octree representation, octree-based transformer with 3D RoPE and token-p arallel generation schemes. OctGPT significantly accelerates convergence, achieves performance rivaling or surpassi...\n\n\nSi-Tong Wei, Rui-Huan Wa ng, Chuan-Zhi Zhou, Baoquan Chen, and Peng-Shuai Wang (Peking University)\ n---------------------\nBANG: Dividing 3D Assets via Generative Exploded D ynamics\n\nBANG introduces Generative Exploded Dynamics, a novel method th at dynamically decomposes 3D objects into meaningful, volumetric parts thr ough smooth, controllable exploded views. Bridging intuitive human underst anding and generative AI, it enables precise part-level manipulation, sema ntic comprehens...\n\n\nLongwen Zhang, Qixuan Zhang, and Haoran Jiang (Sha nghaiTech University, Deemos Technology); Yinuo Bai (ShanghaiTech Universi ty); Wei Yang (Huazhong University of Science and Technology); and Lan Xu and Jingyi Yu (ShanghaiTech University)\n---------------------\nBecoming S pace: Exploring Agential Materiality Through AI-Generated Metamorphosis in Artistic Practice\n\n"Becoming Space" is an installation that explores th e agency of AI, discourse, and material intersections through AI-generated forms and 3D printing. Inspired by Ovid's Metamorphoses, it explores huma n-animal transformations using CLIP-guided diffusion models and stereolith ography. The installation ...\n\n\nXinyu Ma and Hengyu Meng (The Hong Kong University of Science and Technology (Guangzhou)); Ziwei Wu (The Hong Kon g University of Science and Technology); and Zeyu Wang and Clea von Chamie r-Waite (The Hong Kong University of Science and Technology (Guangzhou), T he Hong Kong University of Science and Technology)\n---------------------\ nFloating Strokes: A Spatial Interpretation and Modeling Method of Chinese Calligraphy\n\nTransforming 2D Chinese calligraphy into 3D forms deepens how traditional art is understood and experienced, combining cultural heri tage with modern technology. This approach adds spatial depth, opening new possibilities for digital art, preservation, and interactive design.\n\nB y merging computationa...\n\n\nTroy TianYu LIN (Hong Kong University of Sc ience and Technology, Guangzhou); Boyan Zheng (Independent Researcher); an d Haichuan Lin, Wen You, Kang Zhang, and Chen Liang (Hong Kong University of Science and Technology, Guangzhou)\n---------------------\nwave2weave: A Procedural Weave Data Generation\n\nThis paper presents a procedural dat a generation method for Jacquard weaving that uses matrix computations to create textiles with complex shaded patterns and a triple-layer structure. Employing this method, the authors creatively applied noise functions to weave design and produced a textile artwor...\n\n\nTatsuki Hayama (Keio Un iversity) and Kotaro Uchibe (Hosoo Co.,Ltd)\n---------------------\nDigita l Crazing: Applying Quadtree Structures to Simulate Time and Human Interve ntion\n\nThis work offers an innovative approach to digitally replicating crazing patterns, which are aesthetic crazing found on ceramics. By using a quadtree structure, the method captures the time dependent and user-inte raction aspects of these patterns, providing a novel perspective in digita l material de...\n\n\nSzu-Han Lu and Hsiao-Ching Chou (National Yang Ming Chiao Tung University, Institute of Applied Arts)\n---------------------\n Spherical Lighting with Spherical Harmonics Hessian\n\nWe introduce spheri cal harmonics Hessian and solid spherical harmonics, a variant of spherica l harmonics, to compute the spherical harmonics Hessian efficiently and ac curately to the computer graphics community. These mathematical tools are used to develop an analytical representation of the Hessian...\n\n\nKei Iw asaki (Saitama University, Prometech CG Research) and Yoshinori Dobashi (H okkaido University, Prometech CG Research)\n---------------------\nGuided Lens Sampling for Efficient Monte Carlo Circle-of-Confusion Rendering\n\nA guided lens sampling technique that improves Monte Carlo rendering of dep th-of-field by projecting a global 3D radiance field into lens space via b ipolar-cone projection. This method efficiently targets high-contribution regions, significantly reducing noise and improving convergence for circle -of...\n\n\nJiawei Huang (International Digital Economy Academy), Shaokun Zheng and Kun Xu (Tsinghua University), Yoshifumi Kitamura (Tohoku Univers ity), and Jiaping Wang (International Digital Economy Academy)\n---------- -----------\nBernstein Bounds for Caustics\n\nWe derive vertex position an d irradiance bounds for each triangle tuple, introducing a bounding proper ty of rational functions on the Bernstein basis, to significantly reduce t he search domain when systematically simulating specular light transport.\ n\n\nZhimin Fan, Chen Wang, Yiming Wang, Boxuan Li, and Yuxuan Guo (Nanjin g University); Ling-Qi Yan (University of California Santa Barbara); and Y anwen Guo and Jie Guo (Nanjing University)\n---------------------\nReservo ir Splatting for Temporal Path Resampling and Motion Blur\n\nWe introduce reservoir splatting, a technique preserving exact primary hits during temp oral ReSTIR. This approach makes temporal path resampling more robust unde r motion, especially for regions with high-frequency detail. We further de monstrate how reservoir splatting naturally enables ReSTIR suppor...\n\n\n Jeffrey Liu (University of Illinois Urbana-Champaign); Daqi Lin, Markus Ke ttunen, and Chris Wyman (NVIDIA); and Ravi Ramamoorthi (NVIDIA, University of California San Diego)\n---------------------\nKernel Predicting Neural Shadow Maps\n\nWe present a novel shadow method named kernel predicting n eural shadow mapping. By modeling soft shadow values as pixelwise local fi ltering from basic hard shadow values, we trained a neural network to pred ict local filter weights, achieving accurate and temporally-stable soft sh adows with good gene...\n\n\nXuejun Hu, Jinfan Lu, and Kun Xu (Tsinghua Un iversity)\n---------------------\nStroke Transfer for Participating Media\ n\nWe present a stroke-based method for transforming dynamic 3D scenes wit h smoke, fire, or clouds into painterly animations. Learning from user-pro vided exemplars, our system transfers stroke styles—color, width, length, and orientation—while preserving motion and structure. This enables e...\n \n\nNaoto Shirashima (AGU); Hideki Todo (Takushoku University); Yuki Yamao ka (AGU); Shizuo Kaji (Kyushu University, Kyoto University); and Kunihiko Kobayashi, Haruna Shimotahira, and Yonghao Yue (AGU)\n-------------------- -\nModeling and Rendering Glow Discharge\n\nThis work presents a physicall y-based model for simulating and rendering glow discharge, a luminous plas ma effect seen in neon lights and gas discharge lamps. The model captures particle interactions and emission dynamics, integrates into volume render ing systems, and enables realistic, interactive ...\n\n\nVenkataram Edavam adathil Sivaram, Ravi Ramamoorthi, and Tzu-Mao Li (University of Californi a San Diego)\n---------------------\nPainting with 3D Gaussian Splat Brush es\n\nWe present the first interactive system for painting with 3D Gaussia n splat brushes. With our tool, artists can sample volumetric fragments fr om real-world Gaussian splat captures and paint with them in real time. Ou r tool seamlessly deforms sampled splats along painted strokes, introducin g realisti...\n\n\nKarran Pandey (University of Toronto); Anita Hu, Clemen t Fuji Tsang, and Or Perel (NVIDIA); Karan Singh (University of Toronto); and Maria Shugrina (NVIDIA)\n---------------------\nWhat is HDR? Perceptua l Impact of Luminance and Contrast in Immersive Displays\n\nWe studied pre ferences for different contrasts and peak luminances in HDR. To do this, w e collected a new HDR video dataset, developed tone mappers, and built an HDR haploscope that can reproduce high luminance and contrast. Data was f it to a model which is used for applications like display design...\n\n\nK enneth Chen (New York University; Reality Labs Research, Meta); Nathan Mat suda, Jon McElvain, Yang Zhao, and Thomas Wan (Reality Labs Research, Meta ); Qi Sun (New York University); and Alexandre Chapiro (Reality Labs Resea rch, Meta)\n---------------------\nFaceExpressions-70k: A Dataset of Perce ived Expression Differences\n\nWe introduce FaceExpressions-70k, a large-s cale dataset comprising 70,500 crowdsourced comparisons of facial expressi ons collected from over 1,000 participants. It supports the training of pe rceptual models for expression differences and helps guide decisions on ac ceptable latency and sampling rates...\n\n\nAvinab Saha and Yu-Chih Chen ( University of Texas Austin); Jean-Charles Bazin, Christian Häne, Ioannis K atsavounidis, and Alexandre Chapiro (Reality Labs, Meta); and Alan Bovik ( University of Texas Austin)\n---------------------\nA Deep Learning-based Virtual Oculoplastic Surgery Simulator\n\nA novel deep learning system enh ances realistic virtual oculoplastic surgery simulations.\n\n\nSeonghyeon Kim (KAIST, Visual Media Lab; Anigma Technologies); Chang Wook Seo (Anigma Technologies); Kwanggyoon Seo (KAIST, Visual Media Lab); Seung Han Song ( Chungnam National University Hospital, Chungnam National University); and Junyong Noh (KAIST, Visual Media Lab)\n---------------------\nTowards Unde rstanding Depth Perception in Foveated Rendering\n\nWe demonstrate that st ereoacuity is remarkably resilient to foveated rendering and remains unaff ected with up to 2× stronger foveation than commonly used. To this end, we design a psychovisual experiment and derive a simple perceptual model tha t determines the amount of foveation that does not affec...\n\n\nSophie Ke rgaßner, Taimoor Tariq, and Piotr Didyk (Università della Svizzera italian a)\n---------------------\nQUASAR: Quad-based Adaptive Streaming And Rende ring\n\nThis paper introduces an improved quad-based geometry streaming me thod for remote rendering that reduces bandwidth demands through temporal compression and supports QoE-driven adaptation. It achieves high-quality v isuals, captures disocclusion events, uses 15× less data than SOTA, and re duces bandwi...\n\n\nEdward Lu and Anthony Rowe (Carnegie Mellon Universit y)\n---------------------\nVR-Doh: Hands-on 3D Modeling in Virtual Reality \n\nVR-Doh, an intuitive VR-based 3D modeling system that lets you sculpt and manipulate soft objects and edit 3D Gaussian Splatting scenes in real time. Combining physics-based simulation and expressive interaction, VR-Do h empowers both novices and experts to create rich, deformable, simulation -ready m...\n\n\nZhaofeng Luo (Carnegie Mellon University, Peking Universi ty); Zhitong Cui (Carnegie Mellon University, Zhejiang University); Shijia n Luo (Zhejiang University); Mengyu Chu (Peking University, State Key Labo ratory of General Artificial Intelligence); and Minchen Li (Carnegie Mello n University)\n---------------------\nShape Space Spectra\n\nWe introduce shape-space eigenanalysis to compute eigenfunctions across continuously-pa rameterized shape families. These eigenfunctions are obtained by minimizin g a variational principle. To handle eigenvalue dominance swaps at points of multiplicity, we incorporate dynamic reordering during optimiz...\n\n\n Yue Chang and Otman Benchekroun (University of Toronto), Maurizio M. Chiar amonte (Meta Reality Labs Research), Peter Yichen Chen (MIT CSAIL), and Ei tan Grinspun (University of Toronto)\n---------------------\nDiscrete Tors ion of Connection Forms on Simplicial Meshes\n\nAlthough discrete connecti ons are ubiquitous in vector field design, their torsion remains unstudied . We extend the existing toolbox to control the torsion of discrete connec tions: we introduce a new discrete Levi-Civita connection and define torsi on as a measure of deviation from this reference, so...\n\n\nTheo Braune ( Centre National de la Recherche Scientifique - Laboratoire d'informatique de l'École Polytechnique (LIX), Inria Saclay); Mark Gillespie (Inria Sacla y); Yiying Tong (Michigan State University); and Mathieu Desbrun (Inria Sa clay)\n---------------------\nDesigning 3D Anisotropic Frame Fields with O deco Tensors\n\nOur method proposes a novel computational design framework for designing anisotropic tensor fields. It enables flexible control over scalings without requiring users to specify orientations explicitly. We a pply these anisotropic tensor fields to various applications, such as anis otropic meshing, str...\n\n\nHaikuan Zhu and Hongbo Li (Wayne State Univer sity); Hsueh-Ti Derek Liu (Roblox, University of British Columbia); Wenpin g Wang (Texas A&M University); and Jing Hua and Zichun Zhong (Wayne State University)\n---------------------\nPiecewise Ruled Approximation for Free form Mesh Surfaces\n\nWe propose a method to approximate arbitrary freefor m surface meshes with piecewise ruled surfaces. Our approach optimizes mes h shape and ruling direction field simultaneously, extracts patch topology , and refines ruling positions and orientations. The technique effectively approximates diverse free...\n\n\nYiling Pan, Zhixin Xu, and Bin Wang (Ts inghua University) and Bailin Deng (Cardiff University)\n----------------- ----\nMeschers: Geometry Processing of Impossible Objects\n\nMeschers are a mesh representation for Escheresque geometry. They allow us to solve par tial differential equations on the surface of an impossible object, meanin g that we can find impossible shortest paths, perform mescher smoothing, a nd even inverse render a mescher from an image.\n\n\nAna Dodik, Isabella Y u, Kartik Chandra, and Jonathan Ragan-Kelley (Computer Science and Artific ial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (MIT)); Joshua Tenenbaum (Massachusetts Institute of Technology (MIT)); a nd Vincent Sitzmann and Justin Solomon (Computer Science and Artificial In telligence Laboratory (CSAIL), Massachusetts Institute of Technology (MIT) )\n---------------------\nEnd-to-end Surface Optimization for Light Contro l\n\nDesigning freeform surfaces to reflect or refract light to achieve ta rget light distributions is a challenging inverse problem. We propose an e nd-to-end optimization strategy using a novel differentiable rendering mod el driven by image errors, combined with face-based optimal transport init ializatio...\n\n\nYuou Sun (University of Science and Technology of China) , Bailin Deng (Cardiff University), Juyong Zhang (University of Science an d Technology of China), and Yuou Sun\n---------------------\nPatch-Grid: A n Efficient and Feature-Preserving Neural Implicit Surface Representation\ n\nWe introduce Patch-Grid, a unified neural implicit representation that efficiently represents complex shapes, preserves sharp features, and handl es open boundaries and thin geometric details. By decomposing shapes into patches encapsulated by adaptive feature grids and merging them through lo calized...\n\n\nGuying Lin (Carnegie Mellon University); Lei Yang (The Uni versity of Hong Kong); Congyi Zhang (The University of British Columbia); Hao Pan (Tsinghua University); Yuhan Ping, Guodong Wei, and Taku Komura (T he University of Hong Kong); John Keyser and Wenping Wang (Texas A&M Unive rsity); and Guying Lin\n---------------------\nFast Physics-Based Modeling of Knots and Ties using Templates\n\nWe propose a physics-based modeling system for knots and ties using pipe-like parametric templates, defined by Bézier curves and adaptive radii for flexible, intersection-free shapes. Our method maps cloth regions from UV space into 3D knot forms via a penet ration-free initialization and supports qu...\n\n\nDewen Guo (Peking Unive rsity, Style3D Research); Zhendong Wang and Zegao Liu (Style3D Research); Sheng Li and Guoping Wang (Peking University); Yin Yang (University of Uta h); and Huamin Wang (Style3D Research)\n---------------------\nIntersectio n-Free Garment Retargeting\n\nWe introduce an automatic tool to retarget a rtist-designed garments on a standard mannequin to possibly non-human avat ars with unrealistic characteristics, which widely appear in games and ani mations. We preserve the geometrical features in the original design, guar antee intersection-free, and fit t...\n\n\nZizhou Huang (New York Universi ty, Roblox); Chrystiano Araújo and Andrew Kunz (Roblox); Denis Zorin and D aniele Panozzo (New York University); and Victor Zordan (Roblox, Clemson U niversity)\n---------------------\nRags2Riches: Computational Garment Reus e\n\nWe present the first algorithm to automatically compute sewing patter ns for\nupcycling existing garments into new designs. Our algorithm takes as input\ntwo garment designs along with their corresponding sewing patter ns and\ndetermines how to cut one of them to match the other by following garment\nreus...\n\n\nAnran Qi (INRIA, Université Côte d'Azur); Nico Pietr oni (University of Technology Sydney); Maria Korosteleva (ETH Zurich, Mesh capade); Olga Sorkine-Hornung (ETH Zurich); and Adrien Bousseau (INRIA, Un iversité Côte d'Azur)\n---------------------\nGarmentImage: Raster Encodin g of Garment Sewing Patterns with Diverse Topologies\n\nGarment sewing pat terns often rely on vector formats, which struggle with discontinuities an d unseen topologies. GarmentImage instead encodes geometry, topology, and placement into multi-channel grids, enabling smooth transitions and better generalization. Using simple CNNs, it works well in pattern...\n\n\nYuki Tatsukawa (The University of Tokyo); Anran Qi (INRIA, Université Côte d'Az ur; The University of Tokyo); and I-Chao Shen and Takeo Igarashi (The Univ ersity of Tokyo)\n---------------------\nDress-1-to-3: Single Image to Sim ulation-Ready 3D Outfit with Diffusion Prior and Differentiable Physics\n\ nWe introduce Dress-1-to-3 to reconstruct physics-plausible, simulation-re ady separated garments from an in-the-wild image. Starting with the image, our approach combines a pre-trained image-to-sewing pattern generation mo del with a pre-trained multi-view diffusion model. The sewing pattern is r efine...\n\n\nXuan Li, Chang Yu, Wenxin Du, Ying Jiang, Tianyi Xie, and Yu nuo Chen (University of California Los Angeles); Yin Yang (University of U tah); and Chenfanfu Jiang (University of California Los Angeles)\n-------- -------------\nSingle View Garment Reconstruction Using Diffusion Mapping Via Pattern Coordinates\n\nWe introduce a novel method for accurate 3D gar ment reconstruction from single-view images, bridging 2D and 3D representa tions. Our mapping model creates connections among image pixels, UV coordi nates, and 3D geometry, resulting in realistic garments with intricate det ails and enabling downstream ap...\n\n\nRen Li (EPFL), Cong Cao (MBZUAI), Corentin Dumery and Yingxuan You (EPFL), Hao Li (MBZUAI), and Pascal Fua ( EPFL)\n---------------------\nDiscipline Together with the Self in Kendo: Exploring "Qi" through Mixed Reality and Autoethnography\n\nThis work expl ores “qi” in kendo through mixed reality and autoethnography, blending tra dition and technology. By animating digital humans with “qi”, it frames ma rtial arts as art. The project invites reflection on selfhood and offers f resh insights at the intersection of cul...\n\n\nKaren Furuta, Jingjing Li , Tatsuki Fushimi, and Yoichi Ochiai (University of Tsukuba)\n------------ ---------\nReimagining Beckett’s Not I in Virtual Reality: The MetaHuman a s a Digital Double of the Actor\n\nThis practice-based project reimagines Beckett’s Not I in virtual reality, marrying minimalist theatre with immer sive technology. A lone, disembodied Metahuman mouth exploits VR’s intense presence while subverting customary embodiment and audience agency. Integ rating performing avatars, ...\n\n\nNéill O'Dwyer, Enda Bates, and Nichola s Johnson (Trinity College Dublin)\n---------------------\nPoeSpin: A Huma n-AI Dance to Poetry System for Movement-Based Verse Generation\n\nPoeSpin is a human-AI cocreating system. By transforming pole dance movements int o poetry through AI, we challenge both traditional prejudices against this art form and conventional approaches to human-AI creativity. This work de monstrates how computational systems can preserve the deeply human aspe... \n\n\nYihua Li (Interactive Telecommunications Program, New York Universit y); Hongyue Chen (English Literature and Literary Theory, University of Fr eiburg); Yiqing Li (Interactive Telecommunications Program, New York Unive rsity); and Yetong Xin (Graduate School of Design, Harvard University)\n-- -------------------\nin(A)n(I)mate - AI-Mediated Conversations with Inanim ate Objects\n\n2025, rumored to be the "year of AI agents," the artwork in (A)n(I)mate envisions a future where AI systems act behind the scenes of o bjects, providing them agency and performativity, animating them, and brin ging them closer to human awareness. By inviting conversations with everyd ay objects, in(A)n(...\n\n\nAvital Meshi (University of California Davis) and Adam Wright (UC Davis)\n---------------------\nMulti Layered Autonomy and AI Ecologies in Robotic Art Installations\n\n“Symbiosis of Agents” mer ges AI-driven multi-agent robotics with immersive environments, exploring the delicate balance of machine agency and artist authorship through emerg ent behaviors in self-organized AI ecologies. Its layered approach—micro-l evel strategie, meso-level drives, ...\n\n\nBaoyang Chen (Hong Kong Univer sity of Science and Technology, Central Academy of Fine Arts) and Xian Xu and Huamin Qu (Hong Kong University of Science and Technology)\n---------- -----------\nRenderFormer: Transformer-based Neural Rendering of Triangle Meshes with Global Illumination\n\nWe present RenderFormer, a neural rende ring pipeline that directly renders an image from a triangle-based represe ntation of scene with full global illumination effects, and that does not require per-scene training or finetuning.\n\n\nChong Zeng (State Key Lab o f CAD and CG, Zhejiang University; Microsoft Research Asia); Yue Dong (Mic rosoft Research Asia); Pieter Peers (College of William & Mary); Hongzhi W u (State Key Lab of CAD and CG, Zhejiang University); and Xin Tong (Micros oft Research Asia)\n---------------------\nWishGI: Lightweight Static Glob al Illumination Baking via Spherical Harmonics Fitting\n\nOur work is a li ghtweight static global illumination baking solution that achieves competi tive lighting effects while using only approximately 5% of the memory requ ired by mainstream industry techniques. By adopting a vertex-probe structu re, we ensure excellent runtime performance, making it suitabl...\n\n\nJun ke Zhu (University of Science and Technology of China, Tencent Technology) ; Zehan Wu (Tencent Technology); Qixing Zhang (University of Science and T echnology of China); Cheng Liao (Tencent Technology); and Zhangjin Huang ( University of Science and Technology of China)\n---------------------\nDua l-Band Feature Fusion for Neural Global Illumination with Multi-Frequency Reflections\n\nWe present a neural global illumination method capable of c apturing multi-frequency reflections in dynamic scenes by leveraging objec t-centric feature grids and a novel dual-band fusion module. Our approach produces high-quality, realistic rendering effects and outperforms state-o f-the-art technique...\n\n\nShaohua Mo, Chuankun Zheng, Zihao Lin, and Dia nbing Xi (State Key Lab of CAD&CG, Zhejiang University); Qi Ye (Zhejiang U niversity); Rui Wang and Hujun Bao (State Key Lab of CAD&CG, Zhejiang Univ ersity); and Yuchi Huo (State Key Lab of CAD&CG, Zhejiang University; Zhej iang Lab)\n---------------------\nGaussian Compression for Precomputed Ind irect Illumination\n\nWe propose a Gaussian fitting compression method for light field probes, reducing storage and memory demands in large scenes. Using low-bit adaptive Gaussians and GPU-accelerated decompression, our te chnique replaces traditional PCA-based approaches, achieving 1:50 compress ion ratios. Real-time casc...\n\n\nZhi Zhou (Tencent, University of Scienc e and Technology of China); Chao Li, Zhenyuan Zhang, Mingcong Tang, Zibin Li, and Shuhang Luan (Tencent); and Zhangjin Huang (University of Science and Technology of China)\n---------------------\nSegment-based Light Trans port Simulation\n\nWe introduce a novel segment-based framework for light transport simulation, efficiently assembling paths from disconnected segme nts. Our method includes innovative segment sampling techniques and corres ponding estimation strategies. To demonstrate its strengths, we propose a robust bidirectional pa...\n\n\nWenyou Wang (University of Waterloo), Rex West (Aoyama Gakuin University), and Toshiya Hachisuka (University of Wate rloo)\n---------------------\nNeural Importance Sampling of Many Lights\n\ nNeural approach for estimating spatially varying light selection distribu tions to improve importance sampling in Monte Carlo rendering. To efficien tly manage hundreds or thousands of lights, we integrate our neural approa ch with light hierarchy techniques, where the network predicts cluster-lev el di...\n\n\nPedro Figueiredo and Qihao He (Texas A&M University), Steve Bako (Aurora Innovation), and Nima Khademi Kalantari (Texas A&M University )\n---------------------\nxADA: Controllable and Expressive Audio-Driven A nimation\n\nWe introduce xADA, a generative model for creating expressive and realistic animation of the face, tongue, and head directly from speech audio. \nThe animation maps directly onto MetaHuman compatible rig contro ls enabling integration into industry-standard content creation pipelines. \nxADA generalize...\n\n\nSarah Taylor, Salvador Medina, Jonathan Windle, Erica Alcusa Sáez, and Iain Matthews (Epic Games)\n---------------------\ nDuetGen: Music Driven Two-Person Dance Generation via Hierarchical Masked Modeling\n\nWe present a framework for generating music-driven synchroniz ed two-person dance animations with close interactions. Our system represe nts the two-person motion sequence as a cohesive entity, performs hierarch ical encoding of the motion sequence into discrete tokens, and utilizes du al generative mas...\n\n\nAnindita Ghosh (DFKI, Max Planck Institute for I nformatics); Bing Zhou (Snap Inc.); Rishabh Dabral (Max Planck Institute f or Informatics); Jian Wang (Snap Inc.); Vladislav Golyanik and Christian T heobalt (Max Planck Institute for Informatics); Philipp Slusallek (DFKI, S aarland University); and Chuan Guo (Snap Inc.)\n---------------------\nELG AR: Expressive Cello Performance Motion Generation for Audio Rendition\n\n Generating string instrument performances with intricate movements and com plex interactions poses significant challenges. To address these, we prese nt ELGAR—the first diffusion-based framework for whole-body instrument per formance motion generation solely from audio. We further contribute inno.. .\n\n\nZhiping Qiu and Yitong Jin (Central Conservatory of Music, Tsinghua University); Yuan Wang (Central Conservatory of Music); Yi Shi (Central C onservatory of Music, Tsinghua University); Chao Tan (Weilan Tech); Chongw u Wang, Xiaobing Li, and Feng Yu (Central Conservatory of Music); and Tao Yu and Qionghai Dai (Tsinghua University)\n---------------------\nMotion-e xample-controlled Co-speech Gesture Generation Leveraging Large Language M odels\n\nWe present a framework to utilize Large Language Models (LLMs) fo r co-speech gesture generation with motion examples as direct conditions. It enables multi-modal controls over co-speech gesture generation, such as motion clips, a single pose, human video, or even text prompts.\n\n\nBoho ng Chen (State Key Laboratory of CAD & CG, Zhejiang University) and Yumeng Li, Youyi Zheng, Yao-Xiang Ding, and Kun Zhou (State Key Lab of CAD and C G, Zhejiang University)\n---------------------\nSemantically Consistent Te xt-to-Motion with Unsupervised Styles\n\nWe introduce a novel method that integrates unsupervised style from arbitrary references into a text-driven diffusion model to generate semantically consistent stylized human motion . We leverage text as a mediator to capture the temporal correspondences b etween motion and style, enabling the seamles...\n\n\nLinjun Wu and Xiangj un Tang (Zhejiang University; State Key Laboratory of CAD&CG, Zhejiang Uni versity); Jingyuan Cong (University of California San Diego); He Wang (UCL Centre for Artificial Intelligence, Department of Computer Science, Unive rsity College London (UCL)); Bo Hu, Xu Gong, Songnan Li, and Yuchen Liao ( Tencent Technology (Shenzhen) Co., Ltd.); Yiqian Wu (Zhejiang University); Chen Liu (State Key Lab of CAD and CG, Zhejiang University); and Xiaogang Jin (Zhejiang University; State Key Laboratory of CAD&CG, Zhejiang Univer sity)\n---------------------\nSketch2Anim: Towards Transferring Sketch Sto ryboards into 3D Animation\n\nThis paper presents a novel and first approa ch - Sketch2Anim, to automatically translate 2D storyboard sketches into h igh-quality 3D animations through multi-conditional motion generation.\n\n \nLei Zhong (University of Edinburgh), Chuan Guo (Snap Inc.), Yiming Xie ( Northeastern University), and Jiawei Wang and Changjian Li (University of Edinburgh)\n---------------------\nOn Planar Shape Interpolation With Loga rithmic Metric Blending\n\nLogarithmic metric blending enables smooth inte rpolation between planar shapes while bounding both conformal and area dis tortions. By blending symmetric positive definite metrics in the log domai n, our method geometrically interpolates distortions. This leads to natura l transitions that outperform e...\n\n\nAlon Feldman and Mirela Ben-Chen ( Technion – Israel Institute of Technology)\n---------------------\nClosed- form Generalized Winding Numbers of Rational Parametric Curves for Robust Containment Queries\n\nWe derive closed-form expressions for GWNs of ratio nal parametric curves for robust containment queries. \nOur closed-form ex pression enables efficient computation of GWN, even if the query points ar e located on the rational curve. We also derive the derivatives of GWN for other applications.\n\n\nShibo Liu, Ligang Liu, and Xiao-Ming Fu (Univers ity of Science and Technology of China)\n---------------------\nVariationa l Green and Biharmonic Coordinates for 2D Polynomial Cages\n\nWe present a nalytical formulas for evaluating Green and biharmonic 2D coordinates and their gradients and Hessians, for 2D cages made of polynomial arcs.\nWe pr esent results of 2D image deformations by direct interaction with the cage and through variational solvers.\nWe demonstrate the flexibility\n\n\nEli e Michel, Alec Jacobson, Siddhartha Chaudhuri, and Jean-Marc Thiery (Adobe Research)\n---------------------\nPolynomial 2D Biharmonic Coordinates fo r High-order Cages\n\nWe propose polynomial 2D biharmonic coordinates for closed high-order cages containing polynomial curves of any order by exten ding the classical\n2D biharmonic coordinates using high-order BEM. When a pplying our coordinate\nto 2D cage-based deformation, users manipulate the \Bezier\ncontrol points to q...\n\n\nShibo Liu, Tielin Dai, Ligang Liu, a nd Xiao-Ming Fu (University of Science and Technology of China)\n--------- ------------\nFlexible 3D Cage-based Deformation via Green Coordinates on Bézier Patches\n\nThis work constructs Green coordinates for cages compose d of Bézier patches, which enables flexible deformations with curved bound aries. The high-order structure also allows us to create a compact curved cage for the input models. Additionally, this work proposes a global proje ction technique for pr...\n\n\nDong Xiao and Renjie Chen (University of Sc ience and Technology of China)\n---------------------\nCageNet: A Meta-Fra mework for Learning on Wild Meshes\n\nWe propose a framework for learning on in-the-wild meshes containing non-manifold elements, multiple component s, and interior structures. Our approach uses cages and generalized baryce ntric coordinates to parametrize and learn volumetric functions, demonstra ted by segmentation and skinning weights, ...\n\n\nMichal Edelstein (Techn ion – Israel Institute of Technology); Hsueh-Ti Derek Liu (Roblox, Univers ity of British Columbia); and Mirela Ben-Chen (Technion - Israel Institute of Technology)\n---------------------\nCineMaster: A 3D-Aware and Control lable Framework for Cinematic Text-to-Video Generation\n\nA 3D-aware and c ontrollable text-to-video generation method allows users to manipulate obj ects and camera jointly in 3D space for high-quality cinematic video creat ion.\n\n\nQinghe Wang (Dalian University of Technology); Yawen Luo (The Ch inese University of Hong Kong); Xiaoyu Shi (Kuaishou Technology); Xu Jia a nd Huchuan Lu (Dalian University of Technology); Tianfan Xue (The Chinese University of Hong Kong); and Xintao Wang, Pengfei Wan, Di Zhang, and Kun Gai (Kuaishou Technology)\n---------------------\nMotionCanvas: Cinematic Shot Design with Controllable Image-to-Video Generation\n\nMotionCanvas en ables intuitive cinematic shot design in image-to-video generation by lett ing users control both camera movements and object motions in a 3D-aware s cene. Combining classical graphics with modern diffusion models, it transl ates motion intentions into spatiotemporal signals—withou...\n\n\nJinbo Xi ng (The Chinese University of Hong Kong, Adobe Research); Long Mai, Cusuh Ham, Jiahui Huang, and Aniruddha Mahapatra (Adobe Research); Chi-Wing Fu ( The Chinese University of Hong Kong); Tien-Tsin Wong (Monash University); and Feng Liu (Adobe Research)\n---------------------\nDiffusion as Shader: 3D-aware Video Diffusion for Versatile Video Generation Control\n\nDiffus ion as Shader (DaS) is a unified approach for controlled video generation that uses 3D tracking videos to enable versatile editing, including animat ing mesh-to-video, camera control, motion transfer, and object manipulatio n, while improving temporal consistency.\n\n\nZekai Gu (Hong Kong Universi ty of Science and Technology), Rui Yan (Zhejiang University), Jiahao Lu an d Peng Li (Hong Kong University of Science and Technology), Zhiyang Dou (U niversity of Hong Kong), Chenyang Si (Nanyang Technological University), Z hen Dong (Wuhan University), Qifeng Liu (Hong Kong University of Science a nd Technology), Cheng Lin (University of Hong Kong), Ziwei Liu (Nanyang Te chnological University), Wenping Wang (Texas A&M University), and Yuan Liu (Hong Kong University of Science and Technology)\n---------------------\n Mobius: Text to Seamless Looping Video Generation via Latent Shift\n\nMobi us is a novel method to generate seamlessly looping videos from text descr iptions directly without any user annotations, thereby creating new visual materials for the multi-media presentation.\n\n\nXiuli Bi, Jianfei Yuan, and Bo Liu (Chongqing University of Post and Telecommunications); Yong Zha ng (Meituan); Xiaodong Cun (Great Bay University); Chi-Man Pun (University of Macau); and Bin Xiao (Chongqing University of Post and Telecommunicati ons)\n---------------------\nLayerFlow: A Unified Model for Layer-aware Vi deo Generation\n\nWe propose LayerFlow, a unified framework for layer-awa re video generation, enabling seamless creation of transparent foregrounds , clean backgrounds, and blended scenes. With multi-stage training and LoR A techniques improving layer-wise video quality with limited data, it also supports variants lik...\n\n\nSihui Ji (The University of Hong Kong); Hao Luo (DAMO Academy, Alibaba Group); and Xi Chen, Yuanpeng Tu, Yiyang Wang, and Hengshuang Zhao (The University of Hong Kong)\n---------------------\ nInterspatial Attention for Efficient 4D Human Video Generation\n\nWe intr oduce a novel interspatial attention (ISA) for diffusion transformers, whi ch maintains identity and ensures motion consistency while allowing precis e control of camera and body poses. Combined with a custom video variation autoencoder, our model achieves state-of-the-art performance for photo... \n\n\nRuizhi Shao (Tsinghua University), Yinghao Xu (Stanford University), Yujun Shen (Alibaba Group), Ceyuan Yang (ByteDance Inc.), Yang Zheng and Changan Chen (Stanford University), Yebin Liu (Tsinghua University), and G ordon Wetzstein (Stanford University)\n---------------------\nGenerative V ideo Matting\n\nLimited high-quality ground-truth data hinders traditional video matting's real-world application. This work tackles this by advocat ing for large-scale training with diverse synthetic segmentation and matti ng data. A novel generative pipeline is also introduced to predict tempora lly consistent alpha...\n\n\nYongtao Ge (The University of Adelaide, Zheji ang University); Kangyang Xie, Guangkai Xu, and Mingyu Liu (Zhejiang Unive rsity); Li Ke, Longtao Huang, and Hui Xue (Alibaba Group); Hao Chen (Zheji ang University); and Chunhua Shen (Zhejiang University of Technology, Zhej iang University)\n---------------------\nSpeculative AI Re-enactment of th e Figurists' Encounters With the I Ching\n\nInstead of pursuing the concer n of AI displacing artists, we emphasise a role for artists in reshaping t echnology and branching it in new directions. A role that places us less a s a user of AI technology, waiting for its creative outputs, but as a make r of what AI can be, perhaps leading us towards ...\n\n\nIsaac Joseph Clar ke (Hong Kong University of Science and Technology (Guangzhou)); Raul Masu (Hong Kong University of Science and Technology, Guangzhou; Conservatorio F.A. Bonporti, Italy); and Theo Papatheodorou (Hong Kong University of Sc ience and Technology, Guangzhou)\n---------------------\nAlgorithmic Miner : Humanity in Service - An AI-Driven VR Journey into Machine Logic\n\nAlgo rithmic Miner uses VR to reveal the hidden labor behind AI systems. By imm ersing participants in data annotation tasks, it critically reflects on ex ploitation, automation, and techno-capitalism, prompting new discussions o n ethical, human-centered design in interactive systems.\n\n\nJia SUN (The Hong Kong University of Science and Technology (Guangzhou)); Zheng WEI (T he Hong Kong University of Science and Technology); and Pan HUI (The Hong Kong University of Science and Technology (Guangzhou), The Hong Kong Unive rsity of Science and Technology)\n---------------------\nDigital F(r)ictio ns: Reimagining Colombian Art and its Territory\n\nBoth a critique and cel ebration of digital representation, this project offers multiple perspecti ves beyond technological homogenization. Through exploring digital f(r)ict ions and multiplicities, we reject singular viewpoints in favor of interco nnected truths. Our work with AI and Colombian art rais...\n\n\nAna María Zapata Guzmán, Ludovica Schaerf, Darío Negueruela del Castillo, and Iacopo Neri (University of Zurich, Max Planck Society)\n---------------------\nT alking to the Midnight Broadcast: Reviving 1990s City Memories with AI\n\n This project explores how AI can preserve and reinterpret cultural memory, raising profound questions about the role of technology in connecting pas t and future. By transforming transient, everyday digital interactions int o meaningful archives, it invites reflection on how today’s voices might.. .\n\n\nTongge Yu (Tsinghua University, Massachusetts Institute of Technolo gy (MIT)) and Fan Xiang (Tsinghua University)\n---------------------\nImag e-space Adaptive Sampling for Fast Inverse Rendering\n\nOur goal is to acc elerate inverse rendering by reducing the sampling budget without sacrific ing overall performance. We introduce a novel image-space adaptive samplin g framework to accelerate inverse rendering by dynamically adjusting pixel sampling probabilities based on gradient variance and contr...\n\n\nKai Y an (University of California Irvine); Cheng Zhang (Reality Labs Research, Meta); Sébastien Speierer (Reality Labs, Meta); Guangyan Cai (University o f California Irvine); Yufeng Zhu and Zhao Dong (Reality Labs, Meta); and S huang Zhao (University of California Irvine)\n---------------------\nJames -Stein Gradient Combiner for Inverse Monte Carlo Rendering\n\nThis paper i ntroduces a gradient combiner that blends unbiased and biased gradients in parameter space using the James-Stein estimator to infer scene parameters (BSDFs and volumes) from images. This approach enhances optimization accu racy compared to relying solely on either unbiased or biased gradi...\n\n\ nJeongmin Gu and Bochang Moon (Gwangju Institute of Science and Technology )\n---------------------\nUnbiased Differential Visibility Using Fixed-Ste p Walk-on-Spherical-Caps And Closest Silhouettes\n\nWarped-area reparamete rization is a powerful technique to compute differential visibility. The k ey is constructing a velocity field that is continuous in the domain inter ior and agrees with defined velocities on boundaries. We present a robust and efficient unbiased estimator for differential visibi...\n\n\nLifan Wu, Nathan Morrical, Sai Praveen Bangaru, Rohan Sawhney, Shuang Zhao, Chris W yman, Ravi Ramamoorthi, and Aaron Lefohn (NVIDIA)\n---------------------\n Moment Bounds are Differentiable: Efficiently Approximating Measures in In verse Rendering\n\nMeasures can be compactly represented and approximated using the theory of moments. This work proves that such moment-based repre sentations are differentiable, leading to principled and efficient approac hes for approximating transmittance and visibility in differentiable rende ring.\n\n\nMarkus Worchel and Marc Alexa (TU Berlin)\n-------------------- -\nQuadric-Based Silhouette Sampling for Differentiable Rendering\n\nPhysi cally based differentiable rendering computes gradients of the rendering e quation. The task is made difficult by discontinuities in the integrand at object silhouettes. To address this challenge, we propose a novel edge sa mpling approach that outperforms the state-of-the-art among unidirectiona. ..\n\n\nMariia Soroka (Cornell University, Intel); Christoph Peters (Delft University of Technology, Intel); and Steve Marschner (Cornell University )\n---------------------\nDifferentiable Geometric Acoustic Path Tracing u sing Time-Resolved Path Replay Backpropagation\n\nIntroducing differentiab le path tracing for geometric acoustics with an efficient gradient algorit hm based on path replay backpropagation. The system computes derivatives o f output spectrograms with respect to arbitrary scene parameters (material s, geometry, emitters, microphones) within the framewo...\n\n\nUgo Finnend ahl, Markus Worchel, Tobias Jüterbock, Daniel Wujecki, Fabian Brinkmann, S tefan Weinzierl, and Marc Alexa (TU Berlin)\n---------------------\nStable -Makeup: When Real-World Makeup Transfer Meets Diffusion Model\n\nStable-M akeup is a diffusion-based makeup transfer method. It leverages a Detail-P reserving makeup encoder, and content-structure control modules to preserv e facial content and structure during transfer. Extensive experiments show that Stable-Makeup outperforms existing methods, offering robust, gen...\ n\n\nYuxuan Zhang (Shanghai Jiao Tong University), Yirui Yuan (Shanghai Te ch University), Yiren Song (National University of Singapore), and Jiaming Liu (Tiamat AI)\n---------------------\nFashionComposer: Compositional Fa shion Image Generation\n\nFashionComposer is a flexible model for composit ional fashion image generation, with a universal framework that handles di verse input modalities such as text, human models, and garment images. It personalizes appearance, pose, and human figure, using subject-binding att ention to integrate reference ...\n\n\nSihui Ji, Yiyang Wang, and Xi Chen (The University of Hong Kong); Xiaogang Xu (The Chinese University of Hong Kong); Hao Luo (DAMO Academy, Alibaba Group); and Hengshuang Zhao (The Un iversity of Hong Kong)\n---------------------\nReStyle3D: Scene-Level Appe arance Transfer with Semantic Correspondences\n\nRedesign spaces effortles sly-ReStyle3D transforms indoor scenes by transferring object-specific sty les from a single reference image, preserving 3D coherence. Combining sema ntic-aware diffusion and depth guidance, it enables photo-realistic virtua l staging—faithfully redecorating furniture, te...\n\n\nLiyuan Zhu and She ngqu Cai (Stanford University), Shengyu Huang (NVIDIA), Gordon Wetzstein ( Stanford University), Naji Khosravan (Zillow), and Iro Armeni (Stanford Un iversity)\n---------------------\n3D Stylization via Large Reconstruction Model\n\nGiven a 3D object representing the source content and a reference style image, our method performs 3D stylization with a large pre-trained reconstruction model. This is achieved in a zero-shot manner, with no trai ning or test time optimization required, while delivering superior visual fidelity and ...\n\n\nIpek Oztas (Bilkent University), Duygu Ceylan (Adobe Research), and Aysegul Dundar (Bilkent University)\n--------------------- \nStyle Customization of Text-to-Vector Generation with Image Diffusion Pr iors\n\nWe propose a novel text-to-vector pipeline with style customizatio n that disentangles content and style in SVG generation. Our method repres ents the first feed-forward text-to-vector diffusion model capable of gene rating SVGs in custom styles.\n\n\nPeiying Zhang (City University of Hong Kong), Nanxuan Zhao (Adobe Research), and Jing Liao (City University of Ho ng Kong)\n---------------------\nCobra: Efficient Line Art COlorization wi th BRoAder References\n\nCobra is a novel efficient long-context fine-grai ned ID preservation framework for line art colorization, achieving high pr ecision, efficiency, and flexible usability for comic colorization. By eff ectively integrating extensive contextual references, it transforms black- and-white line art into vibra...\n\n\nJunhao Zhuang (Tsinghua University); Lingen Li, Xuan Ju, and Zhaoyang Zhang (Chinese University of Hong Kong); Chun Yuan (Tsinghua University); and Ying Shan (Tencent)\n--------------- ------\nIMLS-Splatting: Efficient Mesh Reconstruction from Multi-view Imag es via Point Representation\n\nWe propose IMLS-Splatting, an end-to-end mu lti-view mesh optimization method that leverages point clouds for surface representation. By introducing a splatting-based differentiable IMLS algor ithm, our approach efficiently converts point clouds into SDF and texture field, enabling multi-view mesh opt...\n\n\nKaizhi Yang (University of Sci ence and Technology of China); Liu Dai and Isabella Liu (University of Cal ifornia San Diego); Xiaoshuai Zhang (Hillbot Inc.); Xiaoyan Sun and Xuejin Chen (University of Science and Technology of China); Zexiang Xu (Hillbot Inc.); and Hao Su (University of California San Diego, Hillbot Inc.)\n--- ------------------\nSpline Deformation Field\n\nWe combine splines, a cla ssical tool from applied mathematics, with implicit Coordinate Neural Netw orks to model deformation fields, achieving strong performance across mult iple datasets. The explicit regularization from spline interpolation enhan ces spatial coherency in challenging scenarios. We f...\n\n\nMingyang Song (Disney Research Studios, ETH Zürich); Yang Zhang (Disney Research Studio s); Marko Mihajlovic and Siyu Tang (ETH Zürich); Markus Gross (ETH Zürich, Disney Research Studios); and Tunc Ozan Aydin (Disney Research Studios)\n ---------------------\nStochastic Preconditioning for Neural Field Optimiz ation\n\nStochastic preconditioning adds spatial noise to query locations during neural field optimization; it can be formalized as a stochastic est imate for a blur operator. This simple technique eases optimization and si gnificantly improves quality for neural fields optimization, matching or o utperforming ...\n\n\nSelena Ling (NVIDIA, University of Toronto); Merlin Nimier-David (NVIDIA); Alec Jacobson (University of Toronto); and Nicholas Sharp (NVIDIA)\n---------------------\nDiffusing Winding Gradients (DWG): A Parallel and Scalable Method for 3D Reconstruction from Unoriented Poin t Clouds\n\nDiffusing Winding Gradients (DWG) efficiently reconstructs wat ertight 3D surfaces from unoriented point clouds. Unlike conventional meth ods, DWG avoids solving linear systems or optimizing objective functions, enabling simple implementation and parallel execution. Our CUDA implementa tion on an NVIDI...\n\n\nWeizhou Liu (Beijing Normal University); Jiaze Li (Nanyang Technological University); Xuhui Chen and Fei Hou (Institute of Software, Chinese Academy of Sciences; University of Chinese Academy of Sc iences); Shiqing Xin (Shandong University); Xingce Wang and Zhongke Wu (Be ijing Normal University); Chen Qian (SenseTime Group); Ying He (Nanyang Te chnological University College of Computing and Data Science); and Ying H e\n---------------------\nVariational Surface Reconstruction Using Natural Neighbors\n\nWe introduced a new surface reconstruction method from point s without normals. The method robustly handles undersampled regions and sc ales to large input sizes.\n\n\nJianjun Xia and Tao Ju (Washington Univers ity in St. Louis)\n---------------------\nCAST: Component-Aligned 3D Scene Reconstruction from an RGB Image\n\nWe introduce CAST, an innovative meth od for reconstructing high-quality 3D scenes from a single RGB image. Sup porting open-vocabulary reconstruction, CAST excels in managing occlusions , aligning objects accurately, and ensuring physical consistency with the input, unlocking new possibilities in vir...\n\n\nKaixin Yao, Longwen Zhan g, Xinhao Yan, Yan Zeng, and Qixuan Zhang (ShanghaiTech University, Deemos ); Jiayuan Gu (ShanghaiTech University); Wei Yang (Huazhong University of Science and Technology); and Lan Xu and Jingyi Yu (ShanghaiTech University )\n---------------------\nANIME-Rod: Adjustable Nonlinear Isotropic Materi als for Elastic Rods\n\nWe derive a nonlinear elastic rod energy, starting from a general 3D volumetric isotropic material. Validated against FEM, w e accurately capture rod stretching, bending and twisting, under finite de formations. We also propose how to separately control linear/nonlinear str etchability/bendability/twis...\n\n\nHuanyu Chen, Jiahao Wen, and Jernej B arbič (University of Southern California)\n---------------------\nStable C osserat Rods\n\nCosserat rods have become increasingly popular for simulat ing complex thin elastic rods. However, traditional approaches often encou nter significant challenges in robustly and efficiently solving for valid quaternion orientations. We introduce Stable Cosserat rods, which can achi eve high accuracy wi...\n\n\nJerry Hsu (University of Utah), Tongtong Wang and Kui Wu (LightSpeed Studios), and Cem Yuksel (University of Utah)\n--- ------------------\nFast But Accurate: A Real-Time Hyperelastic Simulator with Robust Frictional Contact\n\nThis paper presents a GPU-friendly frame work for real-time implicit simulation of hyperelastic materials with fric tional contacts. Utilizing a novel splitting strategy and efficient solver , the approach achieves robust, high-performance simulation across various stiffness materials, handling large d...\n\n\nZiqiu Zeng (University of S trasbourg; Centre for Artificial Intelligence and Robotics, Hong Kong, CAS ); Siyuan Luo and Fan Shi (National University of Singapore); and Zhongkai Zhang (Centre for Artificial Intelligence and Robotics, Hong Kong, CAS)\n ---------------------\nOptimal r-Adaptive In-Timestep Remeshing for Elasto dynamics\n\nWe propose a coupled mesh-adaptation model and physical simula tion algorithm to jointly generate, per timestep, optimal adaptive remeshi ngs and implicit solutions for the simulation of frictionally contacting, large-deformation elastica.\n\n\nJiahao Wen (University of Southern Califo rnia, Adobe Research); Jernej Barbic (University of Southern California); and Danny Kaufman (Adobe Research)\n---------------------\nVariational Ela stodynamic Simulation\n\nThis paper shows how to express variational time integration for a large class of elastic energies as an optimization probl em with a “hidden” convex substructure. Our integrator improves the perfor mance of elastic simulation tasks, while conserving physical invariants up to tolerance/num...\n\n\nLeticia Mattos Da Silva (Massachusetts Institute of Technology (MIT)); Silvia Sellán (Massachusetts Institute of Technolog y (MIT), Columbia University); and Natalia Pacheco-Tallaj and Justin Solom on (Massachusetts Institute of Technology (MIT))\n---------------------\nA ugmented Vertex Block Descent\n\nWe extend the Vertex Block Descent method for fast and unconditionally stable physics-based simulation using an Aug mented Lagrangian formulation to enable simulating hard constraints with i nfinite stiffness and systems with high stiffness ratios. This allows simu lating complex contact scenarios invo...\n\n\nChris Giles (Roblox) and Eli e Diaz and Cem Yuksel (University of Utah)\n---------------------\nSynedel ica: Mixed Reality Reimagined\n\nSynedelica challenges traditional approac hes to mixed reality by transforming physical environments through a synes thetic experience. This artwork emphasizes the potential for immersive tec hnology to mediate reality itself, fostering social interaction and shared experiences. By reimagining how we p...\n\n\nJohn Desnoyers-Stewart (Univ eristy of British Columbia) and Noah Miller and Bernhard Riecke (Simon Fra ser University)\n---------------------\nHyborg Agency: Fostering AI Agents Through Community Conversations in a Digital Forest\n\nHyborg Agency prop oses an artistic perspective on AI agents: We can design AI agents that ma intain their distinct non-human nature while meaningfully participating in human social contexts.\nPresenting AI agents as mechanical deer nurtured by community conversations, this computational ecosystem demo...\n\n\nYuqi an Sun (Computer Science Research Centre, Royal College of Art); Chenhang Cheng and Chuyan Xu (Individual); Chang Hee Lee (Korea Advanced Institute of Science and Technology (KAIST)); and Ali Asadipour (Computer Science Re search Centre, Royal College of Art)\n---------------------\nLearning to M ove, Learning to Play, Learning to Animate: a Multimedia Exploration of th e More-than-human Intelligence\n\n"Learning to Move, Learning to Play, Lea rning to Animate" is an interdisciplinary multimedia performance, merging real-time AI visuals, plant biofeedback, and found object robotics to expl ore more-than-human intelligence. Challenging anthropocentrism, it envisio ns co-creative agency among humans, ma...\n\n\nMingyong Cheng, Sophia Sun, Han Zhang, and Yuemeng Gu (University of California San Diego)\n--------- ------------\nThe Posthumous World: Our selves, Our Bodies and a World th at Will Continue Without Us\n\nHow would our attitudes to death and dying change if we could see how a human body is reabsorbed into the environment ? The Posthumous World is a project about death and our relationship with the planet. At its centre will be a new artwork - a poetic meditation on a body’s journey to re-join th...\n\n\nRichard Wright (Royal Holloway Unive rsity of London)\n---------------------\nCLR-Wire: Towards Continuous Late nt Representations for 3D Curve Wireframe Generation\n\nCLR-Wire is a unif ied generative framework for 3D curve-based wireframes, jointly modeling g eometry and topology in a continuous latent space. Using attention-driven VAEs and flow matching, it enables high-quality, diverse generation from n oise, images, or point clouds—advancing CAD design, sh...\n\n\nXueqi Ma, Y ilin Liu, Tianlong Gao, Qirui Huang, and Hui Huang (Shenzhen University)\n ---------------------\nPDT: Point Distribution Transformation with Diffusi on Models\n\nPDT is a novel framework that uses diffusion models to transf orm unstructured point clouds into semantically meaningful and structured distributions, such as keypoints, joints, and feature lines. Exploring com plex point distribution transformation, PDT captures fine-grained geometry and semantics, o...\n\n\nJionghao Wang (Texas A&M University); Cheng Lin (University of Hong Kong); Yuan Liu (HKUST); Rui Xu and Zhiyang Dou (Unive rsity of Hong Kong); Xiaoxiao Long (Nanjing University); Haoxiang Guo (Sky work AI, Kunlun Inc.); Taku Komura (University of Hong Kong); and Xin Li a nd Wenping Wang (Texas A&M University)\n---------------------\nRELATE3D: R Efocusing Latent Adapter for Targeted local Enhancement and Editing in 3D Generation\n\nThe alignment of text,images,and 3D is very challenging,yet it is crucial and beneficial for many tasks.We explore and reveal the char acteristics of the native 3D latent space for 3D generation,make it decomp osable and low-rank,thereby enabling efficient learning for multimodal loc al alignment,achie...\n\n\nXiao-Lei Li (Tsinghua University, Tencent Video AI Center); Hao-Xiang Chen (Tsinghua University); Yanni Zhang (Tencent Vi deo AI Center); Kai Ma (Tencent PCG); Alan Zhao (Tencent Video AI Center); Tai-Jiang Mu (Tsinghua University); Haoxiang Guo (Skywork AI, Kunlun Inc. ); and Ran Zhang (Tencent Video AI Center)\n---------------------\nDrag Yo ur Gaussian: Effective Drag-Based Editing with Score Distillation for 3D Gaussian Splatting\n\nDYG is a 3D drag-based scene editing method for Gaus sian Splatting that enables precise, multi-view consistent geometric edits using 3D masks and control points. It combines implicit triplane represen tation and a drag-based diffusion model for high-quality, fine-grained res ults. Visit our project pa...\n\n\nYansong Qu, Dian Chen, and Xinyang Li ( Xiamen University); Xiaofan Li (Baidu Inc.); and Shengchuan Zhang, Liujuan Cao, and Rongrong Ji (Xiamen University)\n---------------------\nCMD: Con trollable Multiview Diffusion for 3D Editing and Progressive Generation\n\ nCMD revolutionizes 3D generation by enabling flexible local editing of 3D models from a single rendering, as well as progressive, interactive creat ion of complex 3D scenes. At its core, CMD leverages a conditional multivi ew diffusion model to seamlessly modify/add new components—enhancing cont. ..\n\n\nPeng Li (Hong Kong University of Science and Technology), Suizhi M a (Johns Hopkins University), Jialiang Chen and Yuan Liu (Hong Kong Univer sity of Science and Technology), Congyi Zhang (Univeristy of British Colum bia), Wei Xue and Wenhan Luo (Hong Kong University of Science and Technolo gy), Alla Sheffer (Univeristy of British Columbia), Wenping Wang (Texas A& M University), and Yike Guo (Hong Kong University of Science and Technolog y)\n---------------------\nSwiftSketch: A Diffusion Model for Image-to-Vec tor Sketch Generation\n\nSwiftSketch, a diffusion-based model with a trans former-decoder, generates high-quality vector sketches from images in unde r a second. It progressively denoises stroke coordinates sampled from a Ga ussian distribution, effectively generalizing across various object classe s. Training uses the ControlS...\n\n\nEllie Arar, Yarden Frenkel, and Dani el Cohen-Or (Tel Aviv University); Ariel Shamir (Reichman University); and Yael Vinker (Computer Science and Artificial Intelligence Laboratory (CSA IL), Massachusetts Institute of Technology (MIT))\n---------------------\n Fluid Simulation on Vortex Particle Flow Maps\n\nWe present the Vortex Par ticle Flow Map (VPFM) method, which revitalizes the traditional Vortex-In- Cell approach for computer graphics. By evolving vorticity and higher-orde r quantities along particle flow maps, our method achieves significantly i mproved long-term stability and vorticity preservatio...\n\n\nSinan Wang ( Georgia Institute of Technology); Junwei Zhou (University of Michigan Ann Arbor, Purdue University); Fan Feng (Dartmouth College); and Zhiqi Li, Yuc hen Sun, Duowen Chen, Greg Turk, and Bo Zhu (Georgia Institute of Technolo gy)\n---------------------\nClebsch Gauge Fluid on Particle Flow Maps\n\nW e present a Clebsch PFM fluid solver that accurately transports wave funct ions using particle flow maps. Key innovations include a new gauge transfo rmation, improved velocity reconstruction on coarse grids, and better fine -scale structure preservation. Benchmarks show superior performance over i mpu...\n\n\nZhiqi Li, Candong Lin, Duowen Chen, and Xinyi Zhou (Georgia In stitute of Technology); Shiying Xiong (Zhejiang University); and Bo Zhu (G eorgia Institute of Technology)\n---------------------\nCirrus: Adaptive H ybrid Particle-Grid Flow Maps on GPU\n\nWe introduce an adaptive octree-ba sed GPU simulator for large-scale fluid simulation. Our hybrid particle-gr id flow map advection scheme effectively preserves vortex details, enablin g high-resolution and high-quality results. The source code has been made publicly available at: https://wang-mengdi.g...\n\n\nMengdi Wang (Georgia Institute of Technology), Fan Feng (Dartmouth College), and Junlin Li and Bo Zhu (Georgia Institute of Technology)\n---------------------\nLeapfrog Flow Maps for Real-Time Fluid Simulation\n\nWe present Leapfrog Flow Maps (LFM), a fast hybrid velocity-impulse scheme with leapfrog integration. Th e computations are further accelerated by a matrix-free AMGPCG solver opti mized for GPUs. As a result, LFM achieves high performance and fidelity ac ross diverse examples, including fireballs and w...\n\n\nYuchen Sun, Junli n Li, Ruicheng Wang, Sinan Wang, and Zhiqi Li (Georgia Institute of Techno logy); Bart G. van Bloemen Waanders (Sandia National Laboratories); and Bo Zhu (Georgia Institute of Technology)\n---------------------\nFluid Simul ation on Compressible Flow Maps\n\nWe present a unified compressible flow map framework based on Lagrangian path integrals, enabling conservative de nsity-energy transport and flexible pressure treatments. Validated on dive rse systems—from shocks to shallow water—it captures complex flow features like vortices and wave int...\n\n\nDuowen Chen and Zhiqi Li (Georgia Inst itute of Technology); Taiyuan Zhang and Jinjin He (Dartmouth College); Jun wei Zhou (University of Michigan, Purdue University); Bart G. van Bloemen Waanders (Sandia National Laboratories); and Bo Zhu (Georgia Institute of Technology)\n---------------------\nEDGE: Epsilon-Difference Gradient Evol ution for Buffer-Free Flow Maps\n\nThis paper presents Epsilon Difference Gradient Evolution (EDGE), a novel method for accurate flow-map computatio n on grids without velocity buffers. EDGE enables large-scale, efficient a nd high-fidelity fluid simulations that capture and preserve complex vorti city structures while significantly red...\n\n\nZhiqi Li, Ruicheng Wang, J unlin Li, Duowen Chen, Sinan Wang, and Bo Zhu (Georgia Institute of Techno logy)\n---------------------\nSobol' Sequences with Guaranteed-Quality 2D Projections\n\nIn the context of quasi-Monte Carlo rendering, we introduce a new Sobol' construction and demonstrate that particular pairs of polyno mials of the form p and p^2+p+1 in Sobol'-based sampling lead to (1, 2)-se quences. They can be combined to form high-dimensional low discrepancy seq uences with good 2D...\n\n\nNicolas Bonneel and David Coeurjolly (CNRS - L IRIS) and Jean-Claude Iehl and Victor Ostromoukhov (Université Claude Bern ard Lyon 1, CNRS - LIRIS)\n---------------------\nCorrect your balance heu ristic: Optimizing balance-style multiple importance sampling weights\n\nM ultiple importance sampling (MIS) is vital to most rendering algorithms. M IS computes a weighted sum of samples from different techniques to handle diverse scene types and lighting effects.\nWe propose a practical weight c orrection scheme that yields better equal-time performance on bidirectiona l al...\n\n\nQingqin Hua and Pascal Grittmann (Saarland University) and Ph ilipp Slusallek (Saarland University, DFKI)\n---------------------\nHistog ram Stratification for Spatio-Temporal Reservoir Sampling\n\nThis paper in troduces stratification into resampled importance sampling (RIS) technique for real-time photorealistic rendering. It organizes sample candidates in to local histograms and then employs Quasi Monte Carlo and antithetic patt erns for efficient sampling. This low-overhead approach significa...\n\n\n Corentin Salaun and Martin Balint (Max Planck Institute for Informatics), Laurent Belcour and Eric Heitz (Intel), and Gurprit Singh and Karol Myszko wski (Max Planck Institute for Informatics)\n---------------------\nMultip le Importance Reweighting for Path Guiding\n\nWe combine the estimates gen erated in each guiding iteration, leveraging the importance distributions from multiple guiding iterations. We demonstrate that our path-level rewei ghting makes guiding algorithms less sensitive to noise and overfitting in distributions.\n\n\nZhimin Fan, Yiming Wang, and Chenxi Zhou (Nanjing Uni versity); Ling-Qi Yan (University of California Santa Barbara); and Yanwen Guo and Jie Guo (Nanjing University)\n---------------------\nVector-Value d Monte Carlo Integration Using Ratio Control Variates\n\nVariance reducti on techniques are widely used to reduce the noise of Monte Carlo integrati on. However, these techniques are typically designed with the assumption t hat the integrand is scalar-valued. To address this, we introduce ratio co ntrol variations, an estimator that leverages a ratio-based ap...\n\n\nHao lin Lu (UC San Diego, MPI for Informatics); Delio Vicini (Google Inc.); an d Wesley Chang and Tzu-Mao Li (UC San Diego)\n---------------------\nPract ical Stylized Nonlinear Monte Carlo Rendering\n\nWe present a practical me thod for rendering scenes with complex, recursive nonlinear stylization ap plied to physically based rendering. Our approach introduces nonlinear pat h filtering(NL-PF) and nonlinear neural radiance caching(NL-NRC), which re duce the exponential sampling cost of stylized render...\n\n\nXiaochun Ton g and Toshiya Hachisuka (University of Waterloo)\n---------------------\nL earning to Assemble with Alternative Plans\n\nWe introduce a reinforcement learning framework for assembling structures composed of rigid parts. A p re-trained policy generates alternative assembly plans, enabling rapid ada ptation to unexpected disruptions. Our approach supports efficient and rob ust planning for multi-robot assembly tasks.\n\n\nZiqi Wang (HKUST, EPFL); Wenjun Liu (HKUST); Jingwen Wang and Gabriel Vallat (EPFL); Fan Shi (Nati onal University of Singapore); and Stefana Parascho and Maryam Kamgarpour (EPFL)\n---------------------\nAMOR: Adaptive Character Control through Mu lti-Objective Reinforcement Learning\n\nPresenting AMOR, a policy conditio ned on context and a linear combination of reward weights, trained using m ulti-objective reinforcement learning. Once trained, AMOR allows for on-th e-fly adjustments of reward weights, unlocking new possibilities in physic s-based and robotic character control.\n\n\nLucas N. Alegre (Instituto de Informática - Universidade Federal do Rio Grande do Sul, Disney Research) and Agon Serifi, Ruben Grandia, David Müller, Espen Knoop, and Moritz Bäch er (Disney Research)\n---------------------\nDesigning Pin-pression Grippe r and Learning its Dexterous Grasping with Online In-hand Adjustment\n\nWe introduce a pin-pression gripper featuring parallel-jaw fingers with 2D a rrays of independently extendable pins, allowing instant shape adaptation to target object geometry and dynamic in-hand re-orientation for enhanced grasp stability. Reinforcement learning with curriculum-based training ena bl...\n\n\nHewen Xiao and Xiuping Liu (Dalian University of Technology), H ang Zhao (Wuhan University), Jian Liu (Shenyang University of Technology), and Kai Xu (National University of Defense Technology (NUDT))\n---------- -----------\nCueTip: An Interactive and Explainable Physics-aware Pool Ass istant\n\nCueTip is an interactive and explainable automated coaching assi stant for a variant of pool/billiards. CueTip has a natural-language inter face, the ability to perform contextual, physics-aware reasoning, and its explanations are rooted in a set of predetermined guidelines developed by domain experts...\n\n\nSean Memery and Kevin Denamganaï (University of Edi nburgh), Jiaxin Zhang and Zehai Tu (Lightspeed Studios), Yiwen Guo (Indepe ndent), and Kartic Subr (University of Edinburgh)\n---------------------\n Creating Fluid-Interactive Virtual Agents by an Efficient Simulator with L ocal-domain Control\n\nWe introduce a novel local-domain fluid-solid inter action simulator grounded in a lattice Boltzmann solver. By leveraging an MPC-based domain-tracking approach and an improved convective boundary con dition, it offers enhanced stability and efficiency for deriving control p olicies of virtual agents, ...\n\n\nWenbin Song, Heng Zhang, Yang Wang, an d Xiaopei Liu (ShanghaiTech University)\n---------------------\nAerial Pat h Online Planning for Urban Scene Updation\n\nWe present the first scene-u pdate aerial path planning algorithm specifically designed for detecting a nd updating change areas in urban environments, which paves the way for ef ficient, scalable, and adaptive UAV-based scene updates in complex urban e nvironments.\n\n\nMingfeng Tang (Shenzhen University), Ningna Wang (Univer sity of Texas at Dallas), Ziyuan Xie (Shenzhen University), Jianwei Hu (Qi Yuan Lab), Ke Xie (Shenzhen University), Xiaohu Guo (University of Texas a t Dallas), and Hui Huang (Shenzhen University)\n---------------------\nMod el See Model Do: Speech-Driven Facial Animation with Style Control\n\nMode lSeeModelDo presents a speech-driven 3D facial animation method using a la tent diffusion model conditioned on a reference clip to capture nuanced pe rformance styles. A novel "style basis" mechanism extracts key poses to gu ide generation, achieving expressive, temporally coherent animations with ...\n\n\nYifang Pan (University of Toronto; Jali Research, Canada); Karan Singh (University of Toronto); and Luiz Gustavo Hafemann (Ubisoft)\n------ ---------------\nTiny is not small enough: High quality, low-resource faci al animation models through hybrid knowledge distillation\n\nThe goal of t his work is to train lip sync animation models that can run in real-time a nd on-device. We design a two-stage knowledge distillation framework to di still large, high-quality models. Our results show that we can train small models with low latency and a comparatively small loss in qualit...\n\n\n Zhen Han, Mattias Teye, Derek Yadgaroff, and Judith Bütepage (Electronic A rts)\n---------------------\nStreamME: Simplify 3D Gaussian Avatar within Live Stream\n\nThe StreamME takes live stream video as input to enable rap id 3D head avatar reconstruction. It achieves impressive speed, capturing the basic facial appearance within 10-seconds and reaching high-quality fi delity within 5-minutes. StreamME reconstructs facial features through on- the-fly training, a...\n\n\nLuchuan Song (University of Rochester, Adobe R esearch); Yang Zhou, Zhan Xu, Yi Zhou, and Deepali Aneja (Adobe Research); and Chenliang Xu (University of Rochester)\n---------------------\nScaffo ldAvatar: High-Fidelity Gaussian Avatars with Patch Expressions\n\nScaffol dAvatar presents a novel approach for reconstructing ultra-high fidelity a nimatable head avatars, which can be rendered in real-time. Our method ope rates on patch-based local expression features and synthesizes 3D Gaussian s dynamically by leveraging tiny scaffold MLPs. We employ color-based d... \n\n\nShivangi Aneja (Technical University of Munich, DisneyResearch|Studi os); Sebastian Weiss, Irene Baeza, Prashanth Chandran, and Gaspard Zoss (D isneyResearch|Studios); Matthias Niessner (Technical University Munich); a nd Derek Bradley (DisneyResearch|Studios)\n---------------------\nSqueezeM e: Mobile-Ready Distillation of Gaussian Full-Body Avatars\n\nExisting Gau ssian Splatting avatars require desktop GPUs, limiting mobile device use. SqueezeMe converts these avatars into a lightweight representation, enabli ng real-time animation and rendering on mobile devices. By distilling the corrective decoder into an efficient linear model, SqueezeMe achie...\n\n\ nForrest Iandola, Stanislav Pidhorskyi, Igor Santesteban, Divam Gupta, Anu j Pahuja, Nemanja Bartolovic, Frank Yu, Emanuel Garbin, Tomas Simon, and S hunsuke Saito (Meta)\n---------------------\nTransforming Unstructured Hai r Strands into Procedural Hair Grooms\n\nRecent methods have been develope d to reconstruct 3D hair strand geometry from images. We introduce an inve rse hair grooming pipeline to transform these unstructured hair strands in to procedural hair grooms controlled by a small set of guide strands and a rtist-friendly grooming operators, enabling e...\n\n\nWesley Chang and And rew Russell (University of California San Diego); Stephane Grabli, Matt Ch iang, Christophe Hery, and Doug Roble (Meta); Ravi Ramamoorthi and Tzu-Mao Li (University of California San Diego); and Olivier Maury (Meta)\n------ ---------------\nBe Decisive: Noise-Induced Layouts for Multi-Subject Gene ration\n\nText-to-image diffusion models struggle with multi-subject gener ation due to subject leakage. Prior methods impose external layouts that c onflict with the model’s prior, harming alignment and natural composition. We introduce a method that leverages the layout encoded in the initial no ise, pro...\n\n\nOmer Dahary (Tel Aviv University, Snap Research); Yehonat han Cohen (Tel Aviv University); Or Patashnik (Tel Aviv University, Snap R esearch); Kfir Aberman (Snap Research); and Daniel Cohen-Or (Tel Aviv Univ ersity, Snap Research)\n---------------------\nCora: Correspondence-aware image editing using few step diffusion\n\nCora is a novel diffusion-based image editing method that achieves complex edits, such as object insertion , background changes, and non-rigid transformations, in only four diffusio n steps. By leveraging pixel-wise semantic correspondences between source and target, it preserves key elements of the o...\n\n\nAmirhossein Alimoha mmadi, Aryan Mikaeili, and Sauradip Nag (Simon Fraser University); Negar H assanpour (Huawei Canada); and Andrea Tagliasacchi and Ali Mahdavi-Amiri ( Simon Fraser University)\n---------------------\nIntrinsicEdit: Precise ge nerative image manipulation in intrinsic space\n\nA generative workflow fo r precise image editing using an intrinsic-image latent space. Built on RG B-X diffusion, it enables diverse edits—like relighting, color changes, an d object manipulation—while preserving identity and ameliorating intrinsic -channel entanglement. All this is done wi...\n\n\nLinjie Lyu (Max-Planck- Institute for Informatics & Saarland Informatics Campus, Adobe Research); Valentin Deschaintre, Yannick Hold-Geoffroy, Milos Hasan, and Jae Shin Yoo n (Adobe Research); Thomas Leimkuehler and Christian Theobalt (Max-Planck- Institute for Informatics & Saarland Informatics Campus); and Iliyan Georg iev (Adobe Research)\n---------------------\n3D-Fixup: Advancing Photo Edi ting with 3D Priors\n\n3D-Fixup enables realistic 3D-aware photo editing b y leveraging 3D priors and a novel data pipeline that extracts training pa irs from real-world videos. Its feed-forward architecture supports efficie nt, high-quality edits involving complex 3D transformations while preservi ng object identity, outperf...\n\n\nYen-Chi Cheng (University of Illinois Urbana-Champaign, Adobe Research); Krishna Kumar Singh and Jae Shin Yoon ( Adobe Research); Alexander Schwing and Liang-Yan Gui (University of Illino is Urbana-Champaign); and Matheus Gadelha, Paul Guerrero, and Nanxuan Zhao (Adobe Research)\n---------------------\nMonetGPT: Solving Puzzles Enhanc es MLLMs’ Image Retouching Skills\n\nMonetGPT explores using multimodal la rge language models (MLLMs) for photo retouching by injecting domain knowl edge via visual puzzles. These puzzles help MLLMs understand individual op erations, visual aesthetics, and generate expert plans. Our procedural pi peline enables explainable edits with det...\n\n\nNiladri Shekhar Dutt (Un iversity College London (UCL)); Duygu Ceylan (Adobe); and Niloy Mitra (Uni versity College London (UCL), Adobe)\n---------------------\nPartEdit: Fin e-Grained Image Editing using Pre-Trained Diffusion Models\n\nWe present P artEdit, a novel diffusion-based system enabling precise, text-based edits of object parts without retraining or manual masks. Optimizing part-aware tokens generates localized non-binary attention maps to guide seamless ed its. Our novel blending strategy delivers high-quality visual resu...\n\n\ nAleksandar Cvejic, Abdelrahman Eldesokey, and Peter Wonka (King Abdullah University of Science and Technology (KAUST))\n---------------------\nInst ance Segmentation of Scene Sketches Using Natural Image Priors\n\nINKi ena bles instance segmentation for scene sketches by adapting image segmentati on models with class-agnostic tuning and depth-based refinement. We introd uce a new dataset INK-scene with diverse styles and demonstrate layered sk etch organization for advanced editing, including inpainting occluded ...\ n\n\nMia Tang (Stanford University, Carnegie Mellon University); Yael Vink er (Computer Science and Artificial Intelligence Laboratory (CSAIL), Massa chusetts Institute of Technology (MIT)); Chuan Yan and Lvmin Zhang (Stanfo rd University); and Maneesh Agrawala (Stanford University, Roblox Research )\n---------------------\nPower-Linear Polar Directional Fields\n\nWe pres ent a method for designing smooth directional fields on triangle meshes wi th precise control over singularities. Our approach uses a power-linear po lar representation, allowing singularities of any index to be placed anywh ere on the mesh. The resulting fields are smooth, robust to mesh qualit... \n\n\nJiabao Brad Wang and Amir Vaxman (University of Edinburgh)\n-------- -------------\nNeurCross: A Neural Approach to Computing Cross Fields for Quad Mesh Generation\n\nWe propose NeurCross, a self-supervised framework for quadrilateral mesh generation that jointly optimizes principal curvatu re direction field and cross field by employing an optimizable neural SDF to approximate the input surface. NeurCross outperforms state-of-the-art m ethods in terms of singular ...\n\n\nQiujie Dong (Shandong University, The University of Hong Kong); Huibiao Wen (Shandong University); Rui Xu (The University of Hong Kong); Shuangmin Chen (Qingdao University of Science an d Technology); Jiaran Zhou (Ocean University of China); Shiqing Xin and Ch anghe Tu (Shandong University); Taku Komura (The University of Hong Kong); and Wenping Wang (Texas A&M University)\n---------------------\nField Smo othness-Controlled Partition for Quadrangulation\n\nOur approach proposes a novel partition method for reliable feature-aligned quadrangulation. The core insight is that singularity-distant smooth streamlines are more suit able as patch boundaries. The key implementation confines patch boundaries to high field smoothness regions.\nValidated on large-sc...\n\n\nZhongxua n Liang, Wei Du, and Xiao-Ming Fu (University of Science and Technology of China)\n---------------------\nFeature-Aligned Parametrization in Penner Coordinates\n\nWe extend Penner-coordinate-based methods for seamless para metrizations to surfaces with sharp features to which the parametrization needs to be aligned. We describe a two-phase method to efficiently minimi ze feature constraint residual errors. We demonstrate that the resulting algorithm works rob...\n\n\nRyan Capouellez and Rodrigo Singh (New York Un iversity), Martin Heistermann and David Bommes (University of Bern), and D enis Zorin (New York University)\n---------------------\nRectangular Surfa ce Parameterization\n\nThis paper presents a method for mapping curved sur faces to the plane without shear, enabling rectangular parameterizations. It introduces a novel approach for computing integrable, orthogonal frame fields. The method improves mesh quality, supports rich user control, and outperforms existing techni...\n\n\nEtienne Corman (CNRS) and Keenan Crane (Carnegie Mellon University)\n---------------------\nDivide-and-Conquer E mbedding\n\nThe paper proposes a construction algorithm based on a divide- and-conquer strategy to map a disk-topology triangular mesh onto any conve x polygon., which supports arbitrary numerical precision and exact arithme tic. Under exact arithmetic, it strictly guarantees a bijection for any me sh and convex po...\n\n\nYuan-Yuan Cheng, Qing Fang, Ligang Liu, and Xiao- Ming Fu (University of Science and Technology of China)\n----------------- ----\n3DGS2: Near Second-order Converging 3D Gaussian Splatting\n\nThis pa per introduces a nearly second-order convergent training algorithm for 3D Gaussian Splatting that exploits independent kernel attributes and sparse coupling across images. By constructing and solving small Newton systems f or parameter groups, it achieves about an-order faster training while m... \n\n\nLei Lan (University of Utah; State Key Lab of CAD and CG, Zhejiang U niversity); Tianjia Shao (Zhejiang University); Zixuan Lu and Yu Zhang (Un iversity of Utah); Chenfanfu Jiang (UCLA); and Yin Yang (University of Uta h)\n---------------------\nFLoD: Integrating Flexible Level of Detail into 3D Gaussian Splatting for Customizable Rendering\n\nFlexible Level of Det ail (FLoD) integrates the concept of LoD into 3DGS using a multi-level rep resentation built with 3D Gaussian scale constraints and level-by-level tr aining strategy. FLoD enables flexible rendering through single-level or s elective rendering for optimal image quality under varyin...\n\n\nYunji Se o, Young Sun Choi, HyunSeung Son, and Youngjung Uh (Yonsei University)\n-- -------------------\nVirtualized 3D Gaussians: Flexible Cluster-based Leve l-of-Detail System for Real-Time Rendering of Composed Scenes\n\nV3DG achi eves real-time rendering of massive 3D Gaussians in large, composed scenes through a novel LOD approach.\nInspired by Nanite, V3DG processes detaile d 3D assets into clusters at various granularities offline, and selectivel y renders 3D Gaussians at runtime—flexibly balancing rendering s...\n\n\nX ijie Yang (Zhejiang University, Shanghai Artificial Intelligence Laborator y); Linning Xu (The Chinese University of Hong Kong); Lihan Jiang (Univers ity of Science and Technology of China, Shanghai Artificial Intelligence L aboratory); Dahua Lin (The Chinese University of Hong Kong, Shanghai Artif icial Intelligence Laboratory); and Bo Dai (University of Hong Kong)\n---- -----------------\nDon’t Splat your Gaussians: Volumetric Ray-Traced Primi tives for Modeling and Rendering Scattering and Emissive Media\n\nWe forma lize the path-tracing of volumes composed of anisotropic kernel mixture mo dels. Our work enables computing physically-based light transport on compl ex volumetric assets efficiently, on tiny memory budgets. We further intro duce Epanechnikov kernels as an efficient alternative in kernel-based ...\ n\n\nJorge Condor (Universita della Svizzera Italiana); Sébastien Speierer , Lukas Bode, Božič Aljaž, and Simon Green (Meta Reality Labs); Piotr Didy k (Universita della Svizzera Italiana); Adrián Jarabo (Meta Reality Labs); and Jorge Condor\n---------------------\nWhen Gaussian Meets Surfel: Ultr a-fast High-fidelity Radiance Field Rendering\n\nWe introduce Gaussian-enh anced Surfels (GESs), a bi-scale representation combining opaque surfels a nd Gaussians for high-fidelity radiance field rendering. GES is entirely s orting free, enabling high-fidelity view-consistent rendering with ultra f ast speeds.\n\n\nKeyang Ye, Tianjia Shao, and Kun Zhou (Zhejiang Universit y)\n---------------------\nDeformable Beta Splatting\n\nDeformable Beta Sp latting (DBS) is a novel approach for real-time radiance field rendering t hat leverages deformable Beta Kernels with adaptive frequency control for both geometry and color encoding. DBS captures complex geometries and ligh ting with state-of-the-art fidelity, while only using 45% fe...\n\n\nRong Liu (USC Institute for Creative Technologies (ICT)), Dylan Sun (University of Southern California), Meida Chen (USC Institute for Creative Technolog ies (ICT)), Yue Wang (University of Southern California), and Andrew Feng (USC Institute for Creative Technologies (ICT))\n---------------------\nIm age-GS: Content-Adaptive Image Representation via 2D Gaussians\n\nWe intro duce Image-GS, a content-adaptive image representation based on colored 2D Gaussians. Image-GS achieves remarkable rate-distortion performance acros s diverse images and textures while supporting hardware-friendly fast rand om access and flexible quality control through a smooth level-of-detai...\ n\n\nYunxiang Zhang and Bingxuan Li (New York University), Alexandr Kuznet sov (Advanced Micro Devices (AMD)), Akshay Jindal and Stavros Diolatzis (I ntel Corporation), Kenneth Chen (New York University), Anton Sochenov and Anton Kaplanyan (Intel Corporation), and Qi Sun (New York University)\n--- ------------------\nBrepDiff: Single-Stage B-rep Diffusion Model\n\nWe pre sent BrepDiff, a simple, single-stage diffusion model for generating Bound ary Representations (B-reps). Our approach generates B-reps by denoising p oint-based face samples with a dedicated noise schedule. Unlike multi-stag e methods, BrepDiff enables intuitive, editable geometry creation, inclu.. .\n\n\nMingi Lee and Dongsu Zhang (Seoul National University), Clément Jam bon (Massachusetts Institute of Technology (MIT)), and Young Min Kim (Seou l National University)\n---------------------\nBoolean Operation for CAD M odels Using a Hybrid Representation\n\nWe propose a novel algorithm for ef ficient and accurate Boolean operations on B-Rep models by mapping them bi jectively to controllable-error triangle meshes. Using conservative inters ection detection on the mesh to locate all surface intersection curves and carefully handling degeneration and topolo...\n\n\nYingyu Yang and Xiaoho ng Jia (State Key Laboratory of Mathematical Sciences, Academy of Mathemat ics and Systems Science, Chinese Academy of Sciences; University of Chines e Academy of Sciences); Bolun Wang (Visual Computing Institute, RWTH Aache n University); Jieyin Yang (State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences; University of Chinese Academy of Sciences); Shiqing Xin (Shandong Universi ty); and Dong-Ming Yan (MAIS, Institute of Automation, Chinese Academy of Sciences; University of Chinese Academy of Sciences)\n-------------------- -\nPS-CAD: Local Geometry Guidance via Prompting and Selection for CAD Rec onstruction\n\nWe propose an iterative prompt-and-select architecture to p rogressively reconstruct the CAD modeling sequence of a target point cloud . We propose the concept of local geometric guidance and come up with thre e ways to integrate this guidance into iterative reconstruction. Experimen ts demonstrate the ...\n\n\nBingchen Yang and Haiyong Jiang (School of Art ificial Intelligence, University of Chinese Academy of Sciences); Hao Pan (Tsinghua University); Guosheng Lin (Nanyang Technological University); Ju n Xiao (School of Artificial Intelligence, University of Chinese Academy o f Sciences); Peter Wonka (KAUST); and Bingchen Yang\n--------------------- \nStitch-A-Shape: Bottom-up Learning for B-Rep Generation\n\nStitch-A-Shap e introduces a novel framework for generating B-Rep models by directly add ressing both topology and geometry. Using a sequential stitching approach, it assembles 3D shapes from vertices through curves to faces, effectively managing topological and geometric complexities. The framework d...\n\n\n Pu Li (Institute of Automation, Chinese Academy Of Sciences) and Wenhao Zh ang, Jinglu Chen, and Dongming Yan (Institute of Automation, Chinese Acade my of Sciences)\n---------------------\nDeFillet: Detection and Removal of Fillet Regions in Polygonal CAD Models\n\nDeFillet, the reverse of CAD fi lleting, is vital for CAE and redesign but challenging with polygon CAD mo dels. Our algorithm uses Voronoi vertices as rolling-ball center candidate s to efficiently identify fillets. Sharp features are then reconstructed v ia quadratic optimization, validated on diverse...\n\n\nJing-En Jiang (Sch ool of Computer Science and Technology, Shandong University); Hanxiao Wang (Institute of Automation, Chinese Academy of Sciences, Beijing, China The School of Artificial Intelligence, University of Chinese Academy of S ciences); Mingyang Zhao (Academy of Mathematics and Systems Science, Chine se Academy of Sciences, the University of Chinese Academy of Sciences); Do ng-Ming Yan (Institute of Automation, Chinese Academy of Sciences); Shuang min Chen (School of Information and Technology, Qingdao University of Scie nce and Technology); Shiqing Xin (School of Computer Science and Technolog y, Shandong University); Changhe Tu (School of Computer Science and Techno logy, Shandong University); and Wenping Wang (Computer Science & Engineer ing, Texas A&M University)\n---------------------\nHoLa: B-Rep Generation using a Holistic Latent Representation\n\nWe introduce a novel representa tion for learning and generating Computer-Aided Design (CAD) models in the form of boundary representations (BReps). Our representation unifies the continuous geometric properties of BRep primitives in different orders (e. g., surfaces and curves) and their\ndiscrete top...\n\n\nYilin Liu, Duoten g Xu, and Xingyao Yu (Shenzhen University); Xiang Xu (Simon Fraser Univers ity, Autodesk Research); Daniel Cohen-Or (Tel Aviv University, Shenzhen Un iversity); Hao Zhang (Simon Fraser University); and Hui Huang (Shenzhen Un iversity)\n---------------------\nPhysically Controllable Relighting of Ph otographs\n\nWe present a photograph relighting method that enables explic it control over light sources akin to CG pipelines. We achieve this in a p ipeline involving mid-level computer vision, physically-based rendering, a nd neural rendering. We introduce a self-supervised training methodology t o train our neura...\n\n\nChris Careaga and Yağız Aksoy (Simon Fraser Univ ersity)\n---------------------\nLightLab: Controlling Light Sources in Ima ges with Diffusion Models\n\nLightLab is a diffusion-based method for para metric control over light sources in an image. Leveraging the linearity of light we create a dataset of controled illumniation changes from a small set of real image pairs and synthetic renders, which is used to fine-tune a model to enable physically plau...\n\n\nNadav Magar (Tel Aviv University , Google); Amir Hertz, Eric Tabellion, Yael Pritch, and Alex Rav-Acha (Goo gle); Ariel Shamir (Reichman University, Google); and Yedid Hoshen (Hebrew University of Jerusalem, Google)\n---------------------\nSpatiotemporally Consistent Indoor Lighting Estimation with Diffusion Priors\n\nWe propose a method for estimating spatiotemporally varying indoor lighting from vid eos using a continuous light field represented as an MLP. By leveraging 2D diffusion priors fine-tuned to predict lighting jointly at multiple locat ions, our approach achieves superior performance and zero-shot gener...\n\ n\nMutian Tong, Rundi Wu, and Changxi Zheng (Columbia University)\n------- --------------\nGSHeadRelight: Fast Relightability for 3D Gaussian Head Sy nthesis\n\nGSHeadRelight enables fast, high-quality relightability for 3D Gaussian head synthesis. A linear light model based on learnable radiance transfer is integrated into the native 3DGS rasterization process and supp orts colored illumination. Without requiring expensive light stage data, o ur method achie...\n\n\nHenglei Lv (Institute of Computing Technology, Chi nese Academy of Sciences; University of Chinese Academy of Sciences); Bail in Deng (Cardiff University); Jianzhu Guo, Xiaoqiang Liu, Pengfei Wan, and Di Zhang (Kuaishou Technology); and Lin Gao (Institute of Computing Techn ology, Chinese Academy of Sciences)\n---------------------\nPractical Inve rse Rendering of Textured and Translucent Appearance\n\nThis work addresse s recovering textured materials using inverse rendering. Our Laplacian mip mapping improves the reconstruction of high-resolution textures. We also p ropose a novel gradient computation that enables efficiently reconstructin g textured, path-traced subsurface scattering. The methods a...\n\n\nPhili ppe Weier (Saarland University, Google); Jérémy Riviere, Ruslan Guseinov, and Stephan Garbin (Google); Philipp Slusallek (Saarland University, DFKI) ; Bernd Bickel (Google, ETH Zürich); and Thabo Beeler and Delio Vicini (Go ogle)\n---------------------\nLayerPano3D: Layered 3D Panorama for Hyper-I mmersive Scene Generation\n\nLayerPano3D is a novel framework that generat es hyper-immersive 3D panoramic scenes from a single text prompt. By decom posing panoramas into multiple layers and optimizing them as 3D Gaussians, it enables full 360°×180° exploration with consistent visual quality, unl ocking new possibilities for virt...\n\n\nShuai Yang (Shanghai Jiao Tong U niversity, Shanghai Artificial Intelligence Laboratory); Jing Tan (The Chi nese University of Hong Kong); Mengchen Zhang (Zhejiang University, Shangh ai Artificial Intelligence Laboratory); Tong Wu (The Chinese University of Hong Kong); Gordon Wetzstein (Stanford University); Ziwei Liu (Nanyang Te chnological University); and Dahua Lin (The Chinese University of Hong Kon g)\n---------------------\nLinear-Time Transport with Rectified Flows\n\nW e propose a simple, parallelizable algorithm inspired by rectified flows t o match probability distributions. With linear-time complexity, it approxi mates optimal transport by employing summed-area tables and direct particl e advection. We illustrate our applications in stippling, mesh parameteriz ati...\n\n\nKhoa Do (University of Michigan), David Coeurjolly (CNRS - LIR IS), Pooran Memari (CNRS - LIX), and Nicolas Bonneel (CNRS - LIRIS)\n----- ----------------\nAccelerated Gamut Discovery via Massive Parallelization\ n\nThis paper proposes a scalable framework using Bayesian Neural Networks and a novel 2mD acquisition function to efficiently discover gamut bounda ries in performance space. Combining NSGA-II's diversity and Bayesian Opti mization's efficiency, the method enables large-batch, parallel optimizati on, out...\n\n\nNavid Ansari, Hans-Peter Seidel, and Vahid Babaei (Max Pla nck Institute for Informatics)\n---------------------\nAdaptive Algebraic Reuse of Reordering in Cholesky Factorizations with Dynamic Sparsity Patte rns\n\nParth delivers adaptive fill-reducing ordering to accelerate Choles ky solvers in simulations with dynamic sparsity patterns, such as contact modelling, achieving up to 255× ordering speedups. With seamless, three-li ne integration into popular solvers like MKL and Accelerate, Parth ensures reliable, ...\n\n\nBehrooz Zarebavani (University of Toronto), Danny M. K aufman (Adobe Research), and David I. W. Levin and Maryam Mehri Dehnavi (U niversity of Toronto)\n---------------------\nMiSo: A DSL for robust and e fficient MINIMIZE and SOLVE problems\n\nMany problems in graphics can be f ormulated as a non-linearly constrained global minimization (MINIMIZE), or solution of a system of non-linear constraints (SOLVE). We introduce MiSo , a domain-specific language and compiler for generating efficient code fo r low-dimensional MINIMIZE and SOLVE problem...\n\n\nFederico Sichetti and Enrico Puppo (Università di Genova), Zizhou Huang (New York University), Marco Attene (CNR IMATI), and Denis Zorin and Daniele Panozzo (New York Un iversity)\n---------------------\nMGPBD: A Multigrid Accelerated Global XP BD Solver\n\nIn high-stiffness, high-resolution simulations, while primal space methods typically fail, the dual-space XPBD method produces unphysic al softening artifacts due to convergence stall. We design an innovative A lgebraic Multigrid method to enhance XPBD, utilizing lazy-update prolongat ors and near-kern...\n\n\nChunlei Li and Peng Yu (Beihang University); Tia ntian Liu (Taichi Graphics); Siyuan Yu (Zenustech); and Yuting Xiao, Shuai Li, Aimin Hao, Yang Gao, and Qinping Zhao (Beihang University)\n--------- ------------\nRigAnything: Template-Free Autoregressive Rigging for Divers e 3D Assets\n\nRigAnything is a transformer-based model that autoregressiv ely generates 3D rigging without templates. It sequentially predicts joint s and skeleton topology while assigning skinning weights, working on objec ts in any pose. It’s 20× faster than existing methods, completing rigging in under 2 se...\n\n\nIsabella Liu (University of California San Diego); Z han Xu, Yifan Wang, and Hao Tan (Adobe Research); Zexiang Xu (Hillbot Inc. ); Xiaolong Wang (University of California San Diego); Hao Su (University of California San Diego, Hillbot Inc.); and Zifan Shi (Adobe Research)\n-- -------------------\nOne Model to Rig Them All: Diverse Skeleton Rigging w ith UniRig\n\nManual 3D rigging is slow. UniRig introduces a unified learn ing framework for automatic skeletal rigging. Trained on our large, divers e Rig-XL dataset, it uses an autoregressive model and cross-attention to a ccurately rig various characters and objects, significantly outperforming prior methods and ...\n\n\nJia-Peng Zhang, Cheng-Feng Pu, and Meng-Hao Guo (CS Dept, Tsinghua University); Yan-Pei Cao (VAST); and Shi-Min Hu (CS De pt, Tsinghua University)\n---------------------\nAnymate: A Dataset and Ba selines for Learning 3D Object Rigging\n\nWe present the Anymate Dataset, a large-scale dataset of 230K 3D assets paired with expert-crafted rigging and skinning information---70 times larger than existing datasets. Using this dataset, we propose a learning-based auto-rigging framework with thre e sequential modules for joint, connectivity, ...\n\n\nYufan Deng, Yuhao Z hang, and Chen Geng (Stanford University); Shangzhe Wu (Stanford Universit y, University of Cambridge); and Jiajun Wu (Stanford University)\n-------- -------------\nMulti-Person Interaction Generation from Two-Person Motion Priors\n\nGenerate exciting multi-character interactions, such as team fig hts, with our training-free method! Multi-character interactions can be de composed into multiple two-person interactions using a directed graph, whi ch enables repurposing large pre-trained two-character motion synthesis mo dels without a...\n\n\nWenning Xu, Shiyu Fan, Paul Henderson, and Edmond S . L. Ho (University of Glasgow)\n---------------------\nLarge-Scale Multi- Character Interaction Synthesis\n\nThis work introduces a conditional gene rative framework for large-scale multi-character interaction synthesis by facilitating natural interactive motions and transitions where characters are coordinated for new interactive partners, proposing a coordinatable mu lti-character interaction space for int...\n\n\nZiyi Chang (Durham Univers ity); He Wang (UCL Centre for Artificial Intelligence, Department of Compu ter Science, University College London (UCL)); and George Koulieris and Hu bert Shum (Durham University)\n---------------------\nSkillMimic-V2: Learn ing Robust and Generalizable Interaction Skills from Sparse and Noisy Demo nstrations\n\nThis work addresses the challenge of learning robust interac tion skills from limited demonstrations. By introducing novel data augment ation techniques for skill transitions and recovery patterns, combined wit h enhanced reinforcement imitation learning methods, we achieve superior p erformance in lear...\n\n\nRunyi Yu (Hong Kong University of Science and T echnology, Shanghai Aritificial Intelligence Laboratory); Yinhuai Wang, Qi han Zhao, and Hok Wai Tsui (Hong Kong University of Science and Technology ); Jingbo Wang (Shanghai Aritificial Intelligence Laboratory); and Ping Ta n and Qifeng Chen (Hong Kong University of Science and Technology)\n------ ---------------\nSplat4D: Diffusion-Enhanced 4D Gaussian Splatting for Tem porally and Spatially Consistent Content Creation\n\nSplat4D generates hig h-fidelity 4D content from monocular videos by integrating multi-view rend ering, inconsistency identification, a video diffusion model, and asymmetr ic U-Net refinement. Our framework maintains spatial-temporal consistency while preserving details and following user guidance, ach...\n\n\nMinghao Yin (University of Hong Kong); Yukang Cao (Nanyang Technological Universit y, Singapore); Songyou Peng (Google DeepMind); and Kai Han (University of Hong Kong)\n---------------------\n4D Gaussian Videos with Motion Layering \n\nWe present 4D Gaussian Video (4DGV) for high-quality, low-storage volu metric video reconstruction and real-time streaming. Our method effectivel y handles complex motion and enables effective motion compression, achievi ng superior performance in both reconstruction quality and storage efficie ncy.\n\n\nPinxuan Dai, Peiquan Zhang, and Zheng Dong (Zhejiang University) ; Ke Xu (City University of Hong Kong); Yifan Peng (The University of Hong Kong); Dandan Ding (Hangzhou Normal University); Yujun Shen (Ant Group); Yin Yang (The University of Utah); Xinguo Liu (Zhejiang University); Rynso n W.H. Lau (City University of Hong Kong); and Weiwei Xu (State Key Lab CA D&CG, Zhejiang University)\n---------------------\nPhotoreal Scene Reconst ruction from an Egocentric Device\n\nThis paper investigates photorealisti c scene reconstruction using videos captured from an egocentric device in high dynamic range. It presents a novel system utilizing visual-inertial b undle adjustment and a physical image formation model that handles camera motion artifacts. The experiments using P...\n\n\nZhaoyang Lv, Maurizio Mo nge, Ka Chen, Yufeng Zhu, Michael Goesele, Jakob Engel, Zhao Dong, and Ric hard Newcombe (Reality Labs Research, Meta)\n---------------------\nOn-the -fly Reconstruction for Large-Scale Novel View Synthesis from Unposed Imag es\n\nWe propose a fast, on-the-fly 3D Gaussian Splatting method that join tly estimates poses and reconstructs scenes. Through fast pose initializat ion, direct primitive sampling, and scalable clustering and merging, it ef ficiently handles diverse ordered image sequences of arbitrary length.\n\n \nAndreas Meuleman, Ishaan Shah, and Alexandre Lanvin (INRIA, Université C ôte d'Azur); Bernhard Kerbl (TU Wien); and George Drettakis (INRIA, Univer sité Côte d'Azur)\n---------------------\nMonocular Online Reconstruction with Enhanced Detail Preservation\n\nWe propose a high-quality online reco nstruction pipeline for monocular input streams, reconstructing environmen ts with detail across multiple levels while maintaining high speed.\n\n\nS ongyin Wu (Meta Reality Labs Research, University of California Santa Barb ara); Zhaoyang Lv, Yufeng Zhu, Duncan Frost, and Zhengqin Li (Meta Reality Labs Research); Ling-Qi Yan (University of California Santa Barbara); and Carl Ren, Richard Newcombe, and Zhao Dong (Meta Reality Labs Research)\n- --------------------\nTransparentGS: Fast Inverse Rendering of Transparent Objects with Gaussians\n\nWe propose TransparentGS, a fast inverse render ing pipeline for transparent objects based on 3D-GS. The main contribution s are three-fold: efficient transparent Gaussian primitives for specular r efraction, GaussProbe to encode ambient light and nearby contents, and the IterQuery algorithm to reduce ...\n\n\nLetian Huang, Dongwei Ye, Jialin D an, and Chengzhi Tao (State Key Lab for Novel Software Technology, Nanjing University); Huiwen Liu (TMCC, College of Computer Science, Nankai Univer sity); Kun Zhou (State Key Lab of CAD&CG, Zhejiang University; Institute o f Hangzhou Holographic Intelligent Technology); Bo Ren (TMCC, College of C omputer Science, Nankai University); and Yuanqi Li, Yanwen Guo, and Jie Gu o (State Key Lab for Novel Software Technology, Nanjing University)\n----- ----------------\npOps: Photo-Inspired Diffusion Operators\n\npOps is a fr amework for learning semantic manipulations in CLIP’s image embedding spac e. Built on a Diffusion Prior model, it enables concept manipulation by tr aining operators directly on image embeddings. This approach enhances sema ntic control and integrates easily with diffusion models for...\n\n\nElad Richardson (Tel Aviv University); Yuval Alaluf (Tel Aviv University, Snap) ; Ali Mahdavi-Amiri (Simon Fraser University); and Daniel Cohen-Or (Tel Av iv University)\n---------------------\nGenerating Past and Future in Digit al Painting Processes\n\nA framework to generate past and future processes for drawing process videos.\n\n\nLvmin Zhang and Chuan Yan (Stanford Univ ersity), Yuwei Guo and Jinbo Xing (CUHK), and Maneesh Agrawala (Stanford U niversity)\n---------------------\nPocket Time-Lapse\n\nPocket Time-Lapse is a system to record, explore and visualize long-term changes in the envi ronment, based on data that a user can capture with the phone they carry. Our contributions include a process to conveniently capture a scene, and n ovel techniques for registering and visualizing panoramic ti...\n\n\nEric Chen (Cornell University; Computer Science and Artificial Intelligence Lab oratory (CSAIL), Massachusetts Institute of Technology (MIT)) and Žiga Kov ačič, Madhav Aggarwal, and Abe Davis (Cornell University)\n--------------- ------\nInstanceGen: Image Generation with Instance-level Instructions\n\n We propose InstanceGen - a new technique for improving Text-to-Image model s ability to generate images for prompts describing multiple objects, attr ibutes and spatial relationships. InstanceGen requires no training or addi tional user inputs and achieves state-of-the art results in terms of both accu...\n\n\nEtai Sella (Tel Aviv University, Meta); Yanir Kleiman (Meta); and Hadar Averbuch-Elor (Cornell Tech)\n---------------------\nIP-Prompte r: Training-Free Theme-Specific Image Generation via Dynamic Visual Prompt ing\n\nThis paper presents T-Prompter, a method for visually prompting gen erative models to enable continuous image generation for specific themes, characters, and scenes. It introduces Dynamic Visual Prompting to enhance generation accuracy and quality, outperforming existing methods in maintai ning charac...\n\n\nYuxin Zhang, Minyan Luo, and Weiming Dong (MAIS, Insti tute of Automation, Chinese Academy of Sciences; School of Artificial Inte lligence, University of Chinese Academy of Sciences); Xiao Yang, Haibin Hu ang, and Chongyang Ma (ByteDance Inc.); Oliver Deussen (University of Kons tanz); Tong-Yee Lee (National Cheng-Kung University); and Changsheng Xu (M AIS, Institute of Automation, Chinese Academy of Sciences; School of Artif icial Intelligence, University of Chinese Academy of Sciences)\n---------- -----------\nDreamMask: Boosting Open-vocabulary Panoptic Segmentation wit h Synthetic Data\n\nTo address a lack of generalization to novel classes, we propose DreamMask, which systematically explores data generation in the open-vocabulary setting, and how to train the model with both real and sy nthetic data. It significantly simplifies the collection of large-scale tr aining data, serving as ...\n\n\nYuanpeng Tu and Xi Chen (The University o f Hong Kong), Ser-Nam Lim (UCF), and Hengshuang Zhao (The University of Ho ng Kong)\n---------------------\nIP-Composer: Semantic Composition of Visu al Concepts\n\nIP-Composer is a novel, training-free method for compositio nal image generation from multiple reference images. Extending IP-Adapter, it uses natural language to identify concept-specific subspaces in CLIP, projects input images into these subspaces to extract targeted concepts, a nd fuses them into ...\n\n\nSara Dorfman and Dana Cohen-Bar (Tel Aviv Univ ersity), Rinon Gal (NVIDIA), and Daniel Cohen-Or (Tel Aviv University)\n-- -------------------\nMIND: Microstructure INverse Design with Generative H ybrid Neural Representation\n\nWe introduce MIND, a novel generative frame work for inverse-designing diverse, tileable 3D microstructures. Leveragin g latent diffusion and our hybrid neural representation, MIND precisely ac hieves targeted physical properties, ensures geometric validity, and enabl es seamless boundary compatibility&...\n\n\nTianyang Xue, Longdu Liu, and Lin Lu (Shandong University); Paul Henderson (University of Glasgow); Peng bin Tang (ETH Zürich); Haochen Li, Jikai Liu, and Haisen Zhao (Shandong Un iversity); Hao Peng (CrownCAD); and Bernd Bickel (ETH Zürich)\n----------- ----------\nNeural Co-Optimization of Structural Topology, Manufacturable Layers, and Path Orientations for Fiber-Reinforced Composites\n\nWe presen t a computational framework that co-optimizes structural topology, curved layers, and fiber orientations for manufacturable, high-strength composite s. Using implicit neural fields, our method integrates design and fabricat ion objectives into a unified optimization process, achieving up to 3...\n \n\nTao Liu, Tianyu Zhang, Yongxue Chen, Weiming Wang, Yu Jiang, Yuming Hu ang, and Charlie C.L. Wang (University of Manchester)\n------------------- --\nData-Efficient Discovery of Hyperelastic TPMS Metamaterials with Extre me Energy Dissipation\n\nWe introduce a method for discovering novel micro scale TPMS structures with high-energy dissipation. By combining a paramet ric design space, empirical testing, and uncertainty-aware deep ensembles with Bayesian optimization, we efficiently explore and discover structures with extreme energy dissipat...\n\n\nMaxine Perroni-Scharf (Massachusetts Institute of Technology (MIT)); Zachary Ferguson (Massachusetts Institute of Technology (MIT), CLO Virtual Fashion); and Thomas Butruille, Carlos P ortela, and Mina Konaković Luković (Massachusetts Institute of Technology (MIT))\n---------------------\nInverse Design of Discrete Interlocking Mat erials with Desired Mechanical Behavior\n\nIn this paper, we present a com putational approach for designing Discrete Interlocking Materials (DIM) wi th desired mechanical properties. We demonstrate the effectiveness of our method by designing discrete interlocking materials with diverse limit pro files for in- and out-of-plane deformation and ...\n\n\nPengbin Tang, Bern hard Thomaszewski, Stelian Coros, and Bernd Bickel (ETH Zürich)\n--------- ------------\nAsymptotic analysis and design of linear elastic shell latti ce metamaterials\n\nThis paper introduces a novel asymptotic directional s tiffness (ADS) metric to analyze the contribution of middle surface geomet ry on the stiffness of shell lattice metamaterials, focusing on Triply Per iodic Minimal Surfaces (TPMS). It provides a theoretical framework and opt imization techniques, ad...\n\n\nDi Zhang and Ligang Liu (University of Sc ience and Technology of China)\n---------------------\nComputational Model ing of Gothic Microarchitecture\n\nGothic microarchitecture—a prevalent fe ature of late medieval art—comprises sculptural works that replicate monum ental Gothic forms, though its original construction techniques remain his torically undocumented. Leveraging insights from 15th-century Basel goldsm ith drawings, we present an...\n\n\nAviv Segall and Jing Ren (ETH Zurich), Martin Schwarz (University of Basel), and Olga Sorkine-Hornung (ETH Zuric h)\n---------------------\nColor Matching and Biomimicry for Multi-materia l Dental 3D Printing\n\nWe propose a practical method for dental layer bio mimicry and multi-spot shade matching using multi-material 3D printing. It integrates seamlessly into workflows combining dental CAD tools and indus trial multi-material slicers. \nWe validated it by printing multiple dentu res and teeth with varying in...\n\n\nAndrás Simon (Fraunhofer IGD, Techni cal University of Darmstadt); Danwu Chen (Fraunhofer IGD); Philipp Urban ( Fraunhofer IGD, Norwegian University of Science and Technology NTNU); and Vincent Duveiller and Henning Lübbe (VITA Zahnfabrik H. Rauter GmbH & Co. KG)\n---------------------\nPolicy-Space Diffusion for Physics-Based Chara cter Animation\n\nWe present a new perspective on physics-based character animation. Assuming policies for similar motions should have similar weigh ts, we introduce regularization during RL training to preserve weight simi larity. By modeling the weights’ manifold with a diffusion model, we gener ate a continuum ...\n\n\nMichele Rocca, Sune Darkner, and Kenny Erleben (U niversity of Copenhagen); Sheldon Andrews (École de Technologie Supérieure (ÉTS)); and Michele Rocca\n---------------------\nMAGNET: Muscle Activati on Generation Networks for Diverse Human Movement\n\nWe introduce MAGNET ( Muscle Activation Generation Networks), a scalable framework for reconstru cting full-body muscle activations across diverse human movements, which a lso includes distilled models for solving downstream tasks or generating r eal-time muscle activations—even on edge devices. T...\n\n\nJungnam Park, Euikyun Jung, Jehee Lee, and Jungdam Won (Seoul National University)\n---- -----------------\nPARC: Physics-based Augmentation with Reinforcement Lea rning for Character Controllers\n\nPARC is a framework that enhances terra in traversal with machine learning and physics-based simulation. By iterat ively training a kinematic motion generator and simulated motion tracker, PARC produces a character controller capable of traversing complex environ ments using highly agile motor skills, ...\n\n\nMichael Xu, Yi Shi, and Ka ngKang Yin (Simon Fraser University) and Xue Bin Peng (Simon Fraser Univer sity, NVIDIA)\n---------------------\nPhysicsFC: Learning User-Controlled Skills for a Physics-Based Football Player Controller\n\nPhysicsFC introdu ces a breakthrough in interactive football simulation—enabling real-time c ontrol of physically simulated players that perform complex skills with sm ooth transitions. It combines skill-specific learning, physics-informed re wards, latent-guided training, and transition-aware sta...\n\n\nMinsu Kim, Eunho Jung, and Yoonsang Lee (Hanyang University)\n---------------------\ nViSA: Physics-based Virtual Stunt Actors for Ballistic Stunts\n\nWe intro duce ViSA (Virtual Stunt Actors), an interactive animation system using de ep reinforcement learning to generate realistic ballistic stunt actions. I t efficiently produces dynamic scenes commonly seen in films and TV dramas , such as traffic accidents and stairway falls. A novel action space d...\ n\n\nMinseok Kim and Wonjeong Seo (Seoul National University), Sung-Hee Le e (Korea Advanced Institute of Science and Technology (KAIST)), and Jungda m Won (Seoul National University)\n---------------------\nPLT: Part-Wise L atent Tokens as Adaptable Motion Priors for Physically Simulated Character s\n\nWe introduce a physically-based character animation framework that ex ploits part-wise latent tokens. The novel structured decomposition enables dynamic exploration to stably adapt to diverse unseen scenarios. Addition al refinement networks improve overall motion quality. We show superior pe rformance...\n\n\nJinseok Bae, Younghwan Lee, Donggeun Lim, and Young Min Kim (Seoul National University)\n---------------------\nDiffuse-CLoC: Guid ed Diffusion for Physics-based Character Look-ahead Control\n\nMeet Diffus e-CLoC—a powerful unification of intuitive steering in kinematic motion ge neration and physics-based character control. By guiding diffusion over jo int state-action spaces, it enables agile, steerable, and physically reali stic motions across diverse downstream tasks—from obsta...\n\n\nXiaoyu Hua ng (University of California Berkeley, Robotics and AI Institute); Takara Truong (Stanford University, Robotics and AI Institute); Yunbo Zhang, Fang zhou Yu, Jean Pierre Sleiman, and Jessica Hodgins (Robotics and AI Institu te); Koushil Sreenath (Robotics and AI Institute, University of California Berkeley); and Farbod Farshidian (Robotics and AI Institute)\n----------- ----------\nMaterialPicker: Multi-Modal DiT-Based Material Generation\n\nM aterialPicker is a multi-modal material generation model that creates high -quality material maps from images and/or text by fine-tuning a video diff usion model. It robustly extracts materials from real-world photos, even w ith distortion or occlusion, enhancing fidelity, diversity, and efficiency in...\n\n\nXiaohe Ma (State Key Lab of CAD&CG, Zhejiang University); Vale ntin Deschaintre, Milos Hasan, and Fujun Luan (Adobe Research); Kun Zhou ( State Key Lab of CAD&CG, Zhejiang University; ZJU-FaceUnity Joint Lab of I ntelligent Graphics); Hongzhi Wu (State Key Lab of CAD&CG, Zhejiang Univer sity); and Yiwei Hu (Adobe Research)\n---------------------\nMatCLIP: Ligh t- and Shape-Insensitive Assignment of PBR Material Models\n\nMatCLIP assi gns realistic PBR materials to 3D models using shape- and lighting-invaria nt descriptors derived from images, including LDM outputs and photos. It o utperforms prior methods by over 15%, enabling consistent material predict ions across varied geometry and lighting, with applications to lar...\n\n\ nMichael Birsak (King Abdullah University of Science and Technology (KAUST )), John Femiani (Miami University), and Biao Zhang and Peter Wonka (King Abdullah University of Science and Technology (KAUST))\n------------------ ---\nPredicting Fabric Appearance Through Thread Scattering and Inversion\ n\nThis paper presents a novel pipeline to digitize physical threads and p redict fabric appearance before fabricating cloth samples, addressing a re al need in the fashion industry. It enables designers to make more informe d material choices, thereby promoting sustainable production, reducing cos ts, and...\n\n\nMengqi Xia, Zhaoyang Zhang, Sumit Chaturvedi, and Yutong Y i (Yale University); Rundong Wu (ByteDance Inc.); and Holly Rushmeier and Julie Dorsey (Yale University)\n---------------------\nPosition-Normal Man ifold for Efficient Glint Rendering on High-Resolution Normal Maps\n\nAccu rate modeling of normal distribution functions (NDF) over a high-resolutio n normal map enables intriguing glinty appearance but is inefficient. We p resent a manifold-based glint formulation, transferring the glint NDF comp utation to mesh intersections. This framework accelerates glint rendering, ...\n\n\nLiwen Wu (University of California San Diego), Fujun Luan and Mil oš Hašan (Adobe Research), and Ravi Ramamoorthi (University of California San Diego)\n---------------------\nA Fluorescent Material Model for Non-Sp ectral Editing & Rendering\n\nWe introduce a material model for diffuse fl uorescence that is compatible with RGB and spectral rendering. This models builds on an analytical integrable Gaussian-based model of the spectral r eradiation that is efficient enough to permits real-time rendering and edi ting of such appearance.\n\n\nLaurent Belcour and Alban Fichet (Intel Labs ) and Pascal Barla (Inria - LaBRI)\n---------------------\nA Fully-statist ical Wave Scattering Model for Heterogeneous Surfaces\n\nThis work present s a statistical wave-scattering model for surfaces with nanoscale mixtures in geometry and material. It predicts average appearance (BRDF) and draws realistic speckles directly from surface statistics, without explicit def initions. The proposed model demonstrates various application...\n\n\nZhen gze Liu and Yuchi Huo (State Key Lab of CAD & CG, Zhejiang University); Yi fan Peng (University of Hong Kong); and Rui Wang (State Key Lab of CAD & C G, Zhejiang University)\n---------------------\nA Fast Parallel Median Fil tering Algorithm Using Hierarchical Tiling\n\nThis paper introduces a nove l median filtering algorithm, using hierarchical tiling to reduce redundan t computations and achieve better complexity than prior sorting-based meth ods. The paper discusses two implementations, for both small and larger ke rnel sizes, that outperform the state of the art b...\n\n\nLouis Sugy (NVI DIA)\n---------------------\nFast Isotropic Median Filtering\n\nThe median filter is a staple of computational image processing. Existing efficient methods share a common flaw, which is that they use a square kernel, produ cing visual artifacts. Our method overcomes this limitation, enabling fast and high-quality circular-kernel median filtering, across multiple ...\n\ n\nBen Weiss (Google Research)\n---------------------\nHigh-Fidelity Novel View Synthesis via Splatting-Guided Diffusion\n\nWe present SplatDiff, a pixel-splatting-guided diffusion model for single-image novel view synthes is (NVS). Leveraging pixel splatting and video diffusion, SplatDiff genera tes high-quality novel views with consistent geometry and high-fidelity de tails. SplatDiff achieves state-of-the-art results in ...\n\n\nXiang Zhang (ETH Zürich, DisneyResearch|Studios); Yang Zhang and Lukas Mehl (DisneyRe search|Studios); Markus Gross (ETH Zürich, DisneyResearch|Studios); and Ch ristopher Schroers (DisneyResearch|Studios)\n---------------------\nCompen sating Spatiotemporally Inconsistent Observations for Online Dynamic 3D Ga ussian Splatting\n\nWe reveal that existing online reconstruction of dynam ic scenes with 3D Gaussian Splatting produces temporally inconsistent resu lts, led by inevitable noise in real-world recordings. To address this, we decompose the rendered images into the ideal signal and the errors during optimization, achieving...\n\n\nYoungsik Yun, Jeongmin Bae, and Hyunseung Son (Yonsei University); Seoha Kim, Hahyun Lee, and Gun Bang (Electronics and Telecommunications Research Institute); and Youngjung Uh (Yonsei Univ ersity)\n---------------------\nDAM-VSR: Disentanglement of Appearance and Motion for Video Super-Resolution\n\nIn this work, we propose DAM-VSR, an appearance and motion disentanglement framework for video super-resolutio n. Appearance enhancement is achieved through reference image super-resolu tion, while motion control is achieved through video ControlNet. Additiona lly, we propose a motion-aligned bidirecti...\n\n\nZhe Kong (Sun Yat-sen U niversity, Meituan); Le Li (Tianjin University); Yong Zhang and Feng Gao ( Meituan); Shaoshu Yang (School of Artificial Intelligence, University of C hinese Academy of Sciences); Tao Wang (Nanjing University); Kaihao Zhang ( Harbin Institute of Technology); Zhuoliang Kang and Xiaoming Wei (Meituan) ; Guanying Chen (Sun Yat-sen University); and Wenhan Luo (The Hong Kong Un iversity of Science and Technology)\n---------------------\nDC-VSR: Spatia lly and Temporally Consistent Video Super-Resolution with Video Diffusion Prior\n\nWe propose DC-VSR, a novel video super-resolution approach based on a video diffusion prior. DC-VSR leverages Spatial and Temporal Attentio n Propagation (SAP and TAP) to ensure spatio-temporally consistent results and Detail-Suppression Self-Attention Guidance (DSSAG) to enhance high-fr equency detai...\n\n\nJanghyeok Han, Gyujin Sim, and Geonung Kim (POSTECH) ; Hyun-Seung Lee, Kyuha Choi, and Youngseok Han (Samsung Electronics); and Sunghyun Cho (POSTECH)\n---------------------\nDynamic Mesh Processing on the GPU\n\nIntroducing the first GPU-based system for dynamic triangle me sh processing, delivering order-of-magnitude speedups over CPU solutions a cross diverse applications. Our system uses patch-based data structure, sp eculative conflict handling, and a novel programming model, enabling robus t, high-performa...\n\n\nAhmed H. Mahmoud (Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (M IT)) and Serban D. Porumbescu and John D. Owens (University of California, Davis)\n---------------------\nStochastic Barnes-Hut Approximation for Fa st Summation on the GPU\n\nWe present a novel stochastic version of the Ba rnes-Hut approximation. Regarding the level-of-detail (LOD) family of appr oximations as control variates, we construct an unbiased estimator of the kernel sum being approximated. Through several examples in graphics, we de monstrate that our method outpe...\n\n\nAbhishek Madan (University of Toro nto); Nicholas Sharp, Francis Williams, and Ken Museth (NVIDIA); and David I.W. Levin (University of Toronto, NVIDIA)\n---------------------\nFast D etermination and Computation of Self-intersections for NURBS Surfaces\n\nD etecting surface self-intersections is crucial for CAD modeling to prevent issues in simulation and manufacturing. This paper presents an algebraic signature-based algorithm for fast determining self-intersections of NURBS surfaces. This signature is then recursively cross-used to compute the se lf-...\n\n\nKai Li and Xiaohong Jia (State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of S ciences; University of Chinese Academy of Sciences); Falai Chen (Universit y of Science and Technology of China); and Kai Li\n---------------------\n Sphere Carving: Bounding Volumes for Signed Distance Fields\n\nWe introduc e Sphere Carving, a method for automatically computing bounding volumes fo r conservative implicit surface. SDF queries define a set of spheres, from which we extract intersection points, used to compute a bounding volume w ith guarantees. Sphere Carving is conceptually simple and independe...\n\n \nHugo Schott (INSA, Lyon; Adobe); Théo Thonat and Thibaud Lambert (Adobe) ; Eric Guérin (INSA, Lyon); Eric Galin (Université Claude Bernard Lyon 1); and Axel Paris (Adobe)\n---------------------\nIn Search of Empty Spheres : 3D Apollonius Diagrams on GPU\n\nWe introduce a novel construction algor ithm of 3D Apollonius diagrams designed for GPUs. Our method features a fa st execution while allowing a comprehensive computation. This is made poss ible thanks to a light data structure, a cell update procedure and a spaci al exploration strategy all designed to...\n\n\nCyprien Plateau--Hollevill e (Université de Limoges, XLIM); Benjamin Stamm (Universität Stuttgart, In stitute of Applied Analysis and Numerical Simulation); Vincent Nivoliers ( Université Claude Bernard Lyon 1, LIRIS); and Maxime Maria and Stéphane Mé rillou (Université de Limoges, XLIM)\n---------------------\nUltraMeshRend erer: Efficient Structure and Management of GPU Out-of-core Memory for Rea l-time Rendering of Gigantic 3D Meshes\n\nThis paper presents UltraMeshRen derer, a GPU out-of-core method for real-time rendering of 3D scenes with billions of vertices and triangles. It features a balanced hierarchical me sh, coherence-based LOD selection, and parallel in-place GPU memory manage ment, achieving efficient data transfer and me...\n\n\nHuadong Zhang (Roch ester Institute of Technology); Lizhou Cao (Rochester Institute of Technol ogy, University of Maryland Eastern Shore); and Chao Peng (Rochester Insti tute of Technology)\n---------------------\nGAIA: Generative Animatable In teractive Avatars with Expression-conditioned Gaussians\n\nWe present GAIA (Generative Animatable Interactive Avatars) for high-fidelity 3D head ava tar generation. GAIA learns dynamic details with expression-conditioned Ga ussians, while being animatable consistently with an underlying morphable model. With a novel two-branch architecture, GAIA disentangles ...\n\n\nZh engming Yu (Texas A&M University); Tianye Li (NVIDIA); Jingxiang Sun (Tsin ghua University, NVIDIA); Omer Shapira, Seonwook Park, Michael Stengel, an d Matthew Chan (NVIDIA); Xin Li and Wenping Wang (Texas A&M University); a nd Koki Nagano and Shalini De Mello (NVIDIA)\n---------------------\nTeGA: Texture Space Gaussian Avatars for High-Resolution Dynamic Head Modeling\ n\nBy combining a continuous, UVD tangent space 3DGS model with a UNet def ormation network while maintaining adaptive densification, we present a no vel high-detail 3D head avatar model that preserves even finer detail like pores and eyelashes at 4K resolution.\n\n\nGengyan S. Li (ETH Zürich, Goo gle) and Paulo Gotardo, Timo Bolkart, Stephan Garbin, Kripasindhu Sarkar, Abhimitra Meka, Alexandros Lattas, and Thabo Beeler (Google)\n------------ ---------\nMyTimeMachine: Personalized Facial Age Transformation\n\nWe per sonalize a pre-trained global aging prior using 50 personal selfies, allow ing age regression (de-aging) and age progression (aging) with high fideli ty and identity preservation.\n\n\nLuchao Qi (University of North Carolina at Chapel Hill (UNC)), Jiaye Wu (University of Maryland College Park), Ba ng Gong (University of North Carolina Chapel Hill), Annie Wang (University of North Carolina at Chapel Hill (UNC)), David Jacobs (University of Mary land College Park), and Roni Sengupta (University of North Carolina at Cha pel Hill (UNC))\n---------------------\nInstantRestore: Single-Step Person alized Face Restoration with Shared-Image Attention\n\nInstantRestore is a fast, personalized face restoration framework that uses a single-step dif fusion model with an extended self-attention mechanism to match low-qualit y image patches to high-quality reference patches. Leveraging implicit cor respondences in the denoising network, we efficiently trans...\n\n\nHoward Zhang (Snap, University of California Los Angeles); Yuval Alaluf (Tel Avi v University); Sizhuo Ma (Snap); Achuta Kadambi (University of California Los Angeles); and Jian Wang and Kfir Aberman (Snap)\n--------------------- \nRelightable Full-Body Gaussian Codec Avatars\n\nWe present the first dri vable full-body avatar model that reconstructs perceptually realistic reli ghtable appearance.\n\n\nShaofei Wang (ETH Zürich); Tomas Simon, Igor Sant esteban, Timur Bagautdinov, Junxuan Li, Vasu Agrawal, Fabian Prada, Shoou- I Yu, Pace Nalbone, Matt Gramlich, Roman Lubachersky, Chenglei Wu, Javier Romero, Jason Saragih, and Michael Zollhoefer (Reality Labs Research, Meta ); Andreas Geiger (University of Tübingen, Tübingen AI Center); Siyu Tang (ETH Zürich); and Shunsuke Saito (Reality Labs Research, Meta)\n---------- -----------\nEVA: Expressive Virtual Avatars from Multi-view Videos\n\nIn this work, we introduce Expressive Virtual Avatars (EVA), an actor-specifi c, fully controllable and expressive human avatar framework that achieves high-fidelity, lifelike renderings in real-time, while enabling independen t control of facial expressions, body movements, and hand gestures.\n\n\nH endrik Junkawitsch, Guoxing Sun, and Heming Zhu (Max Planck Institute for Informatics) and Christian Theobalt and Marc Habermann (Max Planck Institu te for Informatics; Saarbrücken Research Center for Visual Computing, Inte raction and Artificial Intelligence)\n---------------------\nElastic Locom otion with Mixed Second-order Differentiation\n\nOur framework enables rea listic and interesting elastic body locomotion by determining optimal musc le activations to achieve desired movements. It combines interior-point me thod for contact modeling with a novel mixed second-order differentiation algorithm that merges analytic and numerical approach...\n\n\nSiyuan Shen, Tianjia Shao, and Kun Zhou (Zhejiang University); Chenfanfu Jiang (UCLA); Sheldon Andrews (École de Technologie Supérieure (ÉTS)); Victor Zordan (R oblox); and Yin Yang (University of Utah)\n---------------------\nInverse Geometric Locomotion\n\nWe present a computational framework for optimizin g shape sequences to achieve user-defined motion objectives in deformable bodies undergoing geometric locomotion. Through a reduced spatiotemporal p arameterization of the shape sequences, our method is able to efficiently capture the complex coupling...\n\n\nQuentin Becker (EPFL); Oliver Gross ( University of California San Diego, EPFL); and Mark Pauly (EPFL)\n-------- -------------\nImproving Global Motion Estimation in Sparse IMU-based Moti on Capture with Physics\n\nWe propose a physics-driven approach to IMU-bas ed motion capture, improving global motion estimation with 3D contact mode ling and gravity awareness. Our method estimates world-aligned 3D motion, contact points, contact forces, joint torques, and proxy surface interacti ons using only six IMUs in real...\n\n\nXinyu Yi, Shaohua Pan, and Feng Xu (Tsinghua University)\n---------------------\nEncoded Marker Clusters for Auto-Labeling in Optical Motion Capture\n\nMarker-based optical motion ca pture (MoCap) is critical for virtual production and movement sciences. We propose a novel framework for MoCap auto-labeling and matching using uniq uely coded clusters of reflective markers (AEMCs). Compared to commercial software, our method achieves higher labeling ac...\n\n\nHao Wang (Beihang University, Beijing Jiaotong University); Taogang Hou, Tianhui Liu, and J iaxin Li (Beijing Jiaotong University); Tianmiao Wang (Beihang University) ; and Hao Wang and Taogang Hou\n---------------------\nDAMO: A Deep Solver for Arbitrary Marker Configuration in Optical Motion Capture\n\nThis pape r introduces DAMO, a Deep solver for Arbitrary Marker configuration in Opt ical motion capture. DAMO directly infers the relationship between each ra w marker point and 3D model joint, without using predefined marker labels and configuration information.\n\n\nKyeongMin Kim and SeungWon Seo (Korea University), DongHeun Han (KyungHee University), HyeongYeop Kang (Korea Un iversity), and KyeongMin Kim\n---------------------\nHumanRAM: Feed-forwar d Human Reconstruction and Animation Model using Transformers\n\nExisting avatar methods typically require sophisticated dense-view capture and/or t ime-consuming per-subject optimization processes. HumanRAM proposes a feed -forward approach for generalizable human reconstruction and animation fro m monocular or sparse human images. Experiments show that HumanRAM ac...\n \n\nZhiyuan Yu (Department of Mathematics, Hong Kong University of Science and Technology); Zhe Li (Huawei); Hujun Bao (State Key Laboratory of CAD& CG, Zhejiang University); Can Yang (Department of Mathematics, Hong Kong U niversity of Science and Technology); and Xiaowei Zhou (State Key Laborato ry of CAD&CG, Zhejiang Univerisity)\n---------------------\nInstant Self-I ntersection Repair for 3D Meshes\n\nWe present a novel framework that inst antly (< 1 sec) repairs self-intersections in static surface meshes, which commonly occur during the 3D modeling process.\n\n\nWonjong Jang, Yucheol Jung, Gyeongmin Lee, and Seungyong Lee (POSTECH)\n---------------------\n Feature-Preserving Mesh Repair via Restricted Power Diagram\n\nWe present a unified mesh repair framework using a manifold wrap surface to fix diver se imperfections while preserving sharp features. By optimizing projected samples and leveraging adaptive weighting, our method ensures watertightne ss, manifoldness, and high geometric fidelity, outperforming existi...\n\n \nHuibiao Wen (Shandong University, University of Health and Rehabilitatio n Sciences); Guilong He (Shandong University); Rui Xu (University of Hong Kong); Shuangmin Chen (Qingdao University of Science and Technology); Shiq ing Xin (Shandong University); Zhenyu Shu (NingboTech University); Taku Ko mura (University of Hong Kong); Jieqing Feng (State Key Laboratory of CAD & CG, Zhejiang University); Wenping Wang (Texas A&M University); and Chang he Tu (Shandong University)\n---------------------\nTopological Offsets\n\ nTopological Offsets is a method for generating offset surfaces that are t opologically equivalent to an offset infinitesimally close to the surface. By construction, the offsets are manifold, watertight, self-intersection- free, and strictly enclose the input. Tested on Thingi10k, it supports app licat...\n\n\nDaniel Zint, Zhouyuan Chen, Yifei Zhu, and Denis Zorin (New York University/Courant); Teseo Schneider (University of Victoria); and Da niele Panozzo (New York University/Courant)\n---------------------\nSingle Edge Collapse Quad-Dominant Mesh Reduction\n\nA simple and robust modific ation to triangle mesh reduction bridges the gap for what artists want in quad-dominant mesh reduction, preserving symmetry, topology, and joints wi thout sacrificing geometric quality, allowing for high-quality level-of-de tail meshes at no cost compared to what was done be...\n\n\nJulian Knodt ( LightSpeed Studios)\n---------------------\nTetWeave: Isosurface Extractio n using On-The-Fly Delaunay Tetrahedral Grids for Gradient-Based Mesh Opti mization\n\nTetWeave is a novel isosurface representation that jointly opt imizes a tetrahedral grid and directional distances for gradient-based mes h processing like multi-view 3D reconstruction. It dynamically builds adap tive grids via Delaunay triangulation, ensuring watertight, manifold meshe s. By resampling...\n\n\nAlexandre Binninger and Ruben Wiersma (ETH Zurich ), Philipp Herholz (Independent Contributor), and Olga Sorkine-Hornung (ET H Zurich)\n---------------------\nHexHex: Highspeed Extraction of Hexahedr al Meshes\n\nWe present HexHex, which extracts a hexahedral mesh from a lo cally injective integer-grid map. Key contributions include a conservative rasterization technique and a novel mesh data structure called propeller. Our algorithm is significantly faster and uses less memory than the previ ous state-of-the-...\n\n\nTobias Kohler, Martin Heistermann, and David Bom mes (University of Bern)\n---------------------\nUnified Pressure, Surface Tension and Friction for SPH Fluids\n\nWe present a new SPH approach to r eplicate the behavior of droplets and other smaller scale fluid bodies. Fo r this, we develop a new implicit surface tension formulation and implemen t a Coulomb friction force at the fluid-solid interface. A strong coupling between both forces and pressure is achieve...\n\n\nTimo Probst and Matth ias Teschner (University of Freiburg) and Timo Probst\n------------------- --\nA Hybrid Near-wall Model for Kinetic Simulation of Turbulent Boundary Layer Flows\n\nWe present an innovative hybrid near-wall model for the mul ti-resolution lattice Boltzmann solver to effectively enable simulations o f high Reynolds number turbulent boundary layer flows. For the first time, it strikes an excellent balance between the precision demanded by industr ial computational d...\n\n\nMengyun Liu, Kai Bai, and Xiaopei Liu (Shangha iTech University)\n---------------------\nControllable Complex Freezing Dy namics Simulation on Thin Films\n\nWe present a physics-based method for s imulating intricate freezing dynamics on thin films. Our novel Phase Map m ethod integrated with MELP particles reproduces Marangoni freezing dynamic s and the "Snow-Globe Effect". The framework captures soap bubble freezing dynamics while ensuring stability in c...\n\n\nYijie Liu (TMCC, College o f Computer Science, Nankai University); Taiyuan Zhang (Dartmouth College; TMCC, College of Computer Science, Nankai University); and Xiaoxiao Yan, N uoming Liu, and Bo Ren (TMCC, College of Computer Science, Nankai Universi ty)\n---------------------\nDigital Animation of Powder-Snow Avalanches\n\ nPowder-snow avalanches are natural phenomena that result from an instabil ity in the snow cover on a mountain relief. This paper introduces a physic ally-based framework to simulate powder-snow avalanches under complex terr ains, allowing us to animate the turbulent snow cloud dynamics within the avala...\n\n\nFilipe Nascimento, Fabricio S. Sousa, and Afonso Paiva (Univ ersidade de São Paulo - USP)\n---------------------\nArenite: A Physics-ba sed Sandstone Simulator\n\nArenite is a novel, physics-based simulation me thod for generating realistic sandstone structures. It combines fabric int erlocking, multi-factor erosion, and particle-based deposition. Our GPU-ba sed implementation produces detailed 3D shapes such as arches, alcoves, ho odoos, and buttes in minutes an...\n\n\nZhanyu Yang (Purdue University); A ryamaan Jain and Guillaume Cordonnier (Inria, Université Côte d'Azur); Mar ie-Paule Cani (Centre National de la Recherche Scientifique - Laboratoire d'informatique de l'École Polytechnique (LIX), Institut Polytechnique de P aris); and Zhaopeng Wang and Bedrich Benes (Purdue University)\n---------- -----------\nCK-MPM: A Compact-Kernel Material Point Method\n\nWe introduc e a compact, C2-continuous kernel for MPM that reduces numerical diffusion and improves efficiency—without sacrificing stability. Built on a dual-gr id framework and compatible with APIC and MLS, our method enables high-fid elity, large-scale simulations, further pushing the limits of...\n\n\nMich ael Liu (Carnegie Mellon University), Xinlei Wang (NetEase Games Messiah E ngine), and Minchen Li (Carnegie Mellon University)\n--------------------- \nVideoAnydoor: High-fidelity Video Object Insertion with Precise Motion C ontrol\n\nWe propose VideoAnydoor, a zero-shot video object insertion fram ework with high-fidelity detail preservation and precise motion control, w here a pixel warper and a image-video mix-training strategy are designed t o warp the pixel details according to the trajectories. VideoAnydoor demon strates signif...\n\n\nYuanpeng Tu (The University of Hong Kong); Luo Hao (DAMO Academy, Alibaba Group); Chen Xi and Sihui Ji (The University of Hon g Kong); Xiang Bai (Huazhong University of Science and Technology); and Zh ao Hengshuang (The University of Hong Kong)\n---------------------\nSketch 3DVE: Sketch-based 3D-Aware Scene Video Editing\n\nWe propose Sketch3DVE, a sketch-based 3D-aware video editing method to enable detailed local mani pulation of videos with significant viewpoint changes. Our approach levera ges detailed analysis and editing of underlying 3D scene representations, combined with a diffusion model to synthesize realistic...\n\n\nFeng-Lin L iu (Institute of Computing Technology, Chinese Academy of Sciences; Univer sity of Chinese Academy of Sciences); Shi-Yang Li (Institute of Computing Technology, Chinese Academy of Sciences); Yan-Pei Cao (VAST); Hongbo Fu (H ong Kong University of Science and Technology); and Lin Gao (Institute of Computing Technology, Chinese Academy of Sciences; University of Chinese A cademy of Sciences)\n---------------------\nVideoPainter: Any-length Video Inpainting and Editing with Plug-and-Play Context Control\n\nVideoPainter introduces a dual-branch framework for video inpainting with a lightweigh t context encoder that integrates with pre-trained diffusion transformers. Its ID resampling strategy maintains identity consistency across any-leng th videos, while VPData and VPBench provide the largest segmentati...\n\n\ nYuxuan Bian (The Chinese University of Hong Kong, Tencent); Zhaoyang Zhan g (Tencent); Xuan Ju (The Chinese University of Hong Kong); Mingdeng Cao ( The University of Tokyo); Liangbin Xie (University of Macau); Ying Shan (T encent); and Qiang Xu (The Chinese University of Hong Kong)\n------------- --------\nColorSurge: Bringing Vibrancy and Efficiency to Automatic Video Colorization via Dual-Branch Fusion\n\nWe propose ColorSurge, a lightweigh t dual-branch network for end-to-end video colorization. It delivers vivid , accurate, and real-time results from grayscale input, and is easily exte nsible for high-quality performance at low computational cost.\n\n\nHongbo Zhao, Jiaxing Li, Peiyi Zhang, Peng Xiao, Jianxin Lin, and Yijun Wang (Hu nan University)\n---------------------\nControllable Tracking-Based Video Frame Interpolation\n\nWe present a tracking-based video frame interpolati on method, optionally guided by user inputs. It utilizes sparse point trac ks, first estimated using existing point tracking methods and then optiona lly refined by the user. Without any user input, it already achieves state -of-the-art results, with f...\n\n\nKarlis Martins Briedis (DisneyResearch |Studios, ETH Zürich); Abdelaziz Djelouah and Raphaël Ortiz (DisneyResearc h|Studios); Markus Gross (DisneyResearch|Studios, ETH Zürich); and Christo pher Schroers (DisneyResearch|Studios)\n---------------------\nReenact Any thing: Semantic Video Motion Transfer Using Motion-Textual Inversion\n\nRe enact Anything introduces a unified framework for semantic motion transfer , covering applications from full-body and face reenactment to controlling the motion of inanimate objects and the camera. Thereby, motions are repr esented using text/image embeddings of an image-to-video diffusion model a nd...\n\n\nManuel Kansy (ETH Zürich, Disney Research Studios); Jacek Narun iec and Christopher Schroers (Disney Research Studios); Markus Gross (ETH Zürich, Disney Research Studios); and Romann Weber (Disney Research Studio s)\n---------------------\nStressful Tree Modeling: Breaking Branches with Strands\n\nA novel approach for the computational modeling of lignified t issues, such as those found in tree branches and timber, extends strand-ba sed representation to describe biophysical processes at short and long tim e scales. The computationally fast simulation leverages Cosserat rod physi cs and enables t...\n\n\nBosheng Li (Purdue University), Nikolas Schwarz ( Kiel University), Wojtek Palubicki (AMU), Sören Pirk (Kiel University), Do minik L. Michels (King Abdullah University of Science and Technology (KAUS T)), and Bedrich Benes (Purdue University)\n---------------------\nHand-Sh adow Poser\n\nWe solve an inverse hand-shadow problem: finding poses of le ft and right hands that together produce a shadow resembling the target 2D input, e.g., animals, letters, and everyday objects. Our three-stage pipe line decouples the anatomical constraints and semantic constraints, and ou r benchmark provid...\n\n\nHao Xu and Yinqiao Wang (The Chinese University of Hong Kong), Niloy Mitra (University College London (UCL)), Shuaicheng Liu (University of Electronic Science and Technology of China), Pheng Ann Heng (Chinese University of Hong Kong), and Chi-Wing Fu (The Chinese Unive rsity of Hong Kong)\n---------------------\nCurl Quantization for Automati c Placement of Knit Singularities\n\nWe present a method for the automatic placement of knit singularities based on curl quantization. Our method ge nerates knit graphs that maintain all structural manufacturing constraints as well as any additional user constraints. This approach allows for simu lation-free previews of rendered knits an...\n\n\nRahul Mitra (Boston Univ ersity, LightSpeed Studios); Mattéo Couplet (Boston University); Tongtong Wang (LightSpeed Studios); Megan Hoffman (Northeastern University); Kui Wu (LightSpeed Studios); and Edward Chien (Boston University)\n------------- --------\nC-Tubes: Design and Optimization of Tubular Structures Composed of Developable Strips\n\nC-tubes are 3D tubular structures made of develop able strips. We introduce an algorithm to construct C-tubes while guarante eing exact surface developability and an optimization method for design ex ploration. Applications span architecture, engineering, and product design . We present prototypes showc...\n\n\nMichele Vidulis, Klara Mundilova, Qu entin Becker, Florin Isvoranu, and Mark Pauly (EPFL)\n-------------------- -\nDeepMill: Neural Accessibility Learning for Subtractive Manufacturing\n \nThe proposed neural network, DeepMill, can efficiently predict inaccessi ble and occlusion regions in subtractive manufacturing. By utilizing a cut ter-aware dual-head octree-based convolutional architecture, it overcomes the computational inefficiency of traditional geometric methods and is cap able o...\n\n\nFanchao Zhong and Yang Wang (Shandong University), Peng-Shu ai Wang (Peking University), and Lin Lu and Haisen Zhao (Shandong Universi ty)\n---------------------\nDualMS: Implicit Dual-Channel Minimal Surface Optimization for Heat Exchanger Design\n\nDualMS is a novel framework for designing high-performance heat exchangers by directly optimizing the sepa ration surface of two fluids using dual skeleton optimization and neural i mplicit functions. It offers greater topological flexibility than TPMS and achieves superior thermal performance with lo...\n\n\nWeizheng Zhang (Sha ndong University); Hao Pan (School of Software, Tsinghua University); Lin Lu, Xiaowei Duan, and Xin Yan (Shandong University); and Ruonan Wang and Q iang Du (Institute of Engineering Thermophysics, Chinese Academy of Scienc es)\n---------------------\nTowards Comprehensive Neural Materials: Dynami c Structure-Preserving Synthesis with Accurate Silhouette at Instant Infer ence Speed\n\nWe challenge the comprehensive neural material representatio n by thoroughly considering the essential aspects of the complete appearan ce. We introduce an int8-quantized model that keeps high fidelity while ac hieving an order of magnitude speedup compared to previous methods, and a controllable struc...\n\n\nZilin Xu (University of California Santa Barbar a); Xiang Chen (Shandong University); Chen Liu (Zhejiang Lingdi Digital Te chnology Co.,Ltd); Beibei Wang (Nanjing University); Lu Wang (Shandong Uni versity); Zahra Montazeri (University of Manchester); and Ling-Qi Yan (Uni versity of California Santa Barbara)\n---------------------\nGenerative Ne ural Materials\n\nWe present the first generative model for neural BTFs, e nabling single-shot generation from arbitrary text or image prompts. To ac hieve this, we introduce a universal neural material basis and train a con ditional diffusion model to generate materials in this basis from flash im ages, natural images a...\n\n\nNithin Raghavan (University of California S an Diego), Krishna Mullia (Adobe Research), Alexander Trevithick (Universi ty of California San Diego), Fujun Luan and Miloš Hašan (Adobe Research), and Ravi Ramamoorthi (University of California San Diego)\n--------------- ------\nNeural BRDF Importance Sampling by Reparameterization\n\nWe introd uce a reparameterization-based formulation of neural BRDF importance sampl ing. Comparing to previous methods that construct a probability transform to the BRDF through multi-step invertible neural networks, our BRDF sampli ng is in single step without needing network invertibility, achieving...\n \n\nLiwen Wu (University of California San Diego); Sai Bi (Adobe Research) ; Zexiang Xu (Hillbot); Hao Tan, Kai Zhang, and Fujun Luan (Adobe Research ); Haolin Lu (Max Planck Institute for Informatics); and Ravi Ramamoorthi (University of California San Diego)\n---------------------\nRNA: Relighta ble Neural Assets\n\nWe propose a neural representation for 3D assets with complex shading. We precompute shading and scattering on ground-truth geo metry, enabling high-fidelity rendering with full relightability, eliminat ing complex shading models and multiple scattering paths, offering signifi cant speed-ups and seamle...\n\n\nKrishna Mullia, Fujun Luan, Xin Sun, and Miloš Hašan (Adobe Research) and Krishna Mullia\n---------------------\nG enerative detail enhancement for physically based materials\n\nWe present a tool for enhancing the detail of physically based materials using an off -the-shelf diffusion model and inverse rendering. Our goal is to enhance t he visual fidelity of materials with detail that is often tedious to autho r, by adding signs of wear, aging, weathering, etc.\n\n\nSaeed Hadadan (Un iversity of Maryland College Park, NVIDIA); Benedikt Bitterli, Tizian Zelt ner, Jan Novák, Fabrice Rousselle, Jacob Munkberg, Jon Hasselgren, and Bar tlomiej Wronski (NVIDIA); and Matthias Zwicker (University of Maryland Col lege Park)\n---------------------\nAppearance-Preserving Scene Aggregation for Level-of-Detail Rendering\n\nWe present a novel volumetric representa tion for the aggregated appearance of complex scenes and a pipeline for le vel-of-detail generation and rendering. Our representation preserves accur ate far-field appearance and spatial correlation from scene geometry. Our method faithfully reproduces appearanc...\n\n\nYang Zhou and Tao Huang (Un iversity of California Santa Barbara), Ravi Ramamoorthi (University of Cal ifornia San Diego), Pradeep Sen and Ling-Qi Yan (University of California Santa Barbara), and Ling-Qi Yan\n---------------------\nPutting Rigid Bodi es to Rest\n\nWe identify stable orientations of any rigid shape, and the probability that it will rest at these orientations if randomly dropped on the ground. We use a differentiable inverse version of our method to desi gn and fabricate shapes with target resting behavior, such as dice with ta rget, nonuniform p...\n\n\nHossein Baktash (Carnegie Mellon University), N icholas Sharp (NVIDIA), Qingnan Zhou and Alec Jacobson (Adobe Research), a nd Keenan Crane (Carnegie Mellon University)\n---------------------\nA Ver satile Quaternion-Based Constrained Rigid Body Dynamics\n\nWe present an i mplicitly-integrated, quaternion-based constrained Rigid Body Dynamics (RB D) that guarantees satisfaction of kinematic constraints, unifying the sol ution strategy for complex mechanical systems with arbitrary kinematic str uctures, by navigating subspaces spanned by constraint forces a...\n\n\nGu irec Maloisel, Ruben Grandia, Christian Schumacher, Espen Knoop, and Morit z Bächer (Disney Research)\n---------------------\nPainless Differentiable Rotation Dynamics\n\nThis work introduces a forward and differentiable ri gid-body dynamics framework using Lie-algebra rotation derivatives. The ap proach offers simplified, compact derivatives, improved conditioning, and higher efficiency compared to traditional methods. Applications include fu ndamental rigid-body probl...\n\n\nMagí Romanyà (Universidad Rey Juan Carl os) and Miguel A. Otaduy (Universidad Rey Juan Carlos, Meta Reality Labs R esearch)\n---------------------\nGeometric Contact Potential\n\nWe present a systematic derivation of a continuum potential defined for smooth and p iecewise smooth surfaces, by identifying a set of natural requirements for contact potentials. Our potential is formulated independently of surface discretization and addresses the shortcomings of existing potential-...\n\ n\nZizhou Huang and Maxwell Paik (New York University); Zachary Ferguson ( Massachusetts Institute of Technology, CLO Virtual Fashion); and Daniele P anozzo and Denis Zorin (New York University)\n---------------------\nC5D: Sequential Continuous Convex Collision Detection Using Cone Casting\n\nWe propose a fast, single-threaded continuous collision detection (CCD) algor ithm for convex shapes under affine motion. By combining conservative adva ncement with a cone-casting approach, it avoids primitive-level overhead a nd enables efficient integration into intersection-free simulation methods ...\n\n\nXiaodi Yuan (University of California San Diego); Fanbo Xiang (H illbot Inc.); Yin Yang (University of Utah); and Hao Su (University of Cal ifornia San Diego, Hillbot Inc.)\n---------------------\nOffset Geometric Contact\n\nWe introduce Offset Geometric Contact (OGC), a groundbreaking m ethod offering "penetration-free for free" simulations of codimensional ob jects. OGC efficiently constructs offset volumetric shapes to ensure stab le, artifact-free collisions. Leveraging parallel GPU computations, it del ivers real-time...\n\n\nAnka H. Chen (University of Utah, NVIDIA); Jerry H su and Ziheng Liu (University of Utah); Miles Macklin (NVIDIA); and Yin Ya ng and Cem Yuksel (University of Utah)\n---------------------\nElevating 3 D Models: High-Quality Texture and Geometry Refinement from a Low-Quality Model\n\nElevate3D transforms low-quality 3D models into high-quality asse ts through iterative texture and geometry refinement. At its core, HFS-SDE dit refines textures generatively while preserving the input’s identity le veraging high-frequency guidance. The resulting texture then guides geomet ry refi...\n\n\nNuri Ryu, Jiyun Won, Jooeun Son, and Minsu Gong (POSTECH); Joo-Haeng Lee (Pebblous); and Sunghyun Cho (POSTECH)\n------------------- --\nTexture Size Reduction Through Symmetric Overlap and Texture Carving\n \nWe develop a method to compress textures and UVs for meshes in a content -aware way. We combine this with overlapping and folding symmetric UV char ts, and demonstrate our approach on a dataset from Sketchfab. We outperfor m prior work in visual similarity to the original mesh.\n\n\nJulian Knodt and Xifeng Gao (LightSpeed Studios) and Julian Knodt\n-------------------- -\nAlignTex: Pixel-Precise Texture Generation from Multi-view Artwork\n\nA lignTex is a novel framework for generating high-quality textures from 3D meshes and multi-view artwork. It improves texture generation by ensuring both appearance detail and geometric consistency, outpacing traditional me thods in quality and efficiency, making it a valuable tool for 3D asset cr eat...\n\n\nYuqing Zhang, Hao Xu, and Yiqian Wu (Zhejiang University; Stat e Key Laboratory of CAD&CG, Zhejiang University); Sirui Chen and Sirui Lin (Zhejiang University); Xiang Li (Shenzhen University); Xifeng Gao (Lights peed Studios, Tencent America); and Xiaogang Jin (Zhejiang University; Sta te Key Laboratory of CAD&CG, Zhejiang University)\n---------------------\n The Mokume Dataset and Inverse Modeling of Solid Wood Textures\n\nWe prese nt the Mokume dataset for solid wood texturing, comprising nearly 190 samp les from various species. Using this dataset, we propose an inverse modeli ng pipeline to infer volumetric wood textures from surface photographs, em ploying inverse procedural texturing and neural cellular automata (NCA...\ n\n\nMaria Larsson (The University of Tokyo); Hodaka Yamaguchi (Gifu Prefe cture Research Institute for Human Life Technology, Nihon University); Ehs an Pajouheshgar (École Polytechnique Féderale de Lausanne (EPFL)); I-Chao Shen and Kenji Tojo (The University of Tokyo); Chia-Ming Chang (National T aiwan University of Arts); Lars Hansson (Luleå University of Technology, N orwegian University of Science and Technology); Olof Broman (Luleå Univers ity of Technology); Takashi Ijiri (Shibaura Institute of Technology); Arie l Shamir (Reichman University); Wenzel Jakob (The University of Tokyo, Éco le Polytechnique Féderale de Lausanne (EPFL)); and Takeo Igarashi (The Uni versity of Tokyo)\n---------------------\nEscher Tile Deformation via Clos ed-Form Solution\n\nA real-time deformation method for Escher tiles --- in terlocking organic forms that seamlessly tessellate the plane following sy mmetry rules. Rather than treating tiles as mere boundaries, we consider t hem as textured shapes, ensuring that both the boundary and interior defor m simultaneously. The de...\n\n\nCrane He Chen (Industrial Light & Magic, Northeastern University) and Vladimir Kim (Adobe)\n---------------------\n Claycode: Stylable and Deformable 2D Scannable Codes\n\nWe introduce a nov el scannable 2D code where the payload is stored in the topology of nested color regions, abandoning traditional matrix-based approaches (e.g., QRCo des). Claycodes can be largely deformed, styled, and animated. We present a mapping between bits and topologies, shape-constrained ren...\n\n\nMarco Maida, Alberto Crescini, Marco Perronet, and Elena Camuffo (Independent R esearcher)\n---------------------\nSynchronized tracing of primitive-based implicit volumes\n\nThe paper presents a tile-based rendering pipeline fo r modeling with implicit volumes, using blobtrees and smooth CSG operators . It requires no preprocessing when updating primitives and ensures effici ent ray processing with sphere tracing. The method uses a low-resolution A -buffer and bottom-up tre...\n\n\nCédric Zanni (Université de Lorraine CNR S, Inria, LORIA) and Cédric Zanni\n---------------------\nPaRas: A Rasteri zer for Large-Scale Parametric Surfaces\n\nHigher-order surfaces enable co mpact, smooth geometry but require efficient rendering. We introduce PaRas , a GPU-based rasterizer that directly renders parametric surfaces, avoidi ng costly tessellation. It integrates seamlessly into existing pipelines, outperforming traditional methods for quartic t...\n\n\nKechun Wang and Re njie Chen (University of Science and Technology of China)\n--------------- ------\nAppearance-aware Multi-view SVBRDF Reconstruction via Deep Reinfor cement Learning\n\nThis paper introduces an appearance-aware adaptive samp ling method using deep reinforcement learning to optimize the reconstructi on of spatially-varying BRDFs from minimal images. By modeling the samplin g as a sequential decision-making problem, the method identifies the next best view-lighting pair...\n\n\nPengfei Zhu, Jie Guo, Yifan Liu, Qi Sun, Y anxiang Wang, and Keheng Xu (Nanjing University); Ligang Liu (University o f Science and Technology of China); and Yanwen Guo (Nanjing University)\n- --------------------\nUncertainty for SVBRDF Acquisition using Frequency A nalysis\n\nWe quantify uncertainty for SVBRDF acquisition from multi-view captures using entropy. The otherwise heavy computation is accelerated in the frequency domain, yielding a practical, efficient method. We apply unc ertainty to improve SVBRDF capture by guiding camera placement, inpainting uncertain regi...\n\n\nRuben Wiersma (ETH Zürich); Julien Philip (Netflix Eyeline Studios); Miloš Hašan, Krishna Mullia, and Fujun Luan (Adobe Rese arch); Elmar Eisemann (Delft University of Technology); and Valentin Desch aintre (Adobe Research)\n---------------------\nStreaming-Aware Neural Mon te Carlo Rendering Framework with Unified Denoising-Compression and Client Collaboration\n\nTo reduce the high rendering costs and transmission band width requirements of path tracing-based cloud rendering, we propose a nov el streaming-aware rendering framework that is able to learn a joint optim al model integrating two path-tracing acceleration (adaptive sampling and denoising) and video c...\n\n\nHangming Fan, Yuchi Huo, Chuankun Zheng, an d Chonghao Hu (State Key Laboratory of CAD & CG, Zhejiang University); Yaz hen Yuan (Game Engine Department, CROS, Tencent, China); and Rui Wang (Sta te Key Laboratory of CAD & CG, Zhejiang University)\n--------------------- \nFeeling Blue or Seeing Red? Investigating the effect of light color, sha dow and realism on the perception of emotion of real and virtual humans\n\ nThis study explores how light color influences the perception of emotion of virtual characters. By analyzing various lighting conditions, including red and blue hues, we reveal how light affects emotion intensity, recogni tion, and genuineness. Findings show that lighting, realism, and shadows a re ke...\n\n\nRachel McDonnell and Bharat Vyas (Trinity College Dublin), U ros Sikimic (Epic Games), and Pisut Wisessing (CMKL University)\n--------- ------------\nPrimitiveAnything: Human-Crafted 3D Primitive Assembly Gener ation with Auto-Regressive transformer\n\nWe present PrimitiveAnything, a novel framework that reformulates shape primitive abstraction as a primiti ve assembly generation task. PrimitiveAnything can generate 3D high-qualit y primitive assemblies that better align with human perception while maint aining geometric fidelity across diverse shape...\n\n\nJingwen Ye (Tencent AIPD); Yuze He (Tencent AIPD, Tsinghua University); Yanning Zhou, Yiqin Z hu, and Kaiwen Xiao (Tencent AIPD); Yong-Jin Liu (Tsinghua University); an d Wei Yang and Xiao Han (Tencent AIPD)\n---------------------\nImage-Space Collage and Packing with Differentiable Rendering\n\nThis work introduces an efficient image-space collage technique that optimizes geometric layou ts using a differential renderer and hierarchical resolution strategy. Our approach simplifies complex shape handling in image-space optimization, o ffering fixed computational complexity. Experiments show o...\n\n\nZhenyu Wang and Min Lu (Shenzhen University)\n---------------------\nBuildingBloc k: A Hybrid Approach for Structured Building Generation\n\nWe propose Buil dingBlock, a hybrid approach integrating generative models, PCG, and LLMs for diverse and structured 3D building generation. A Transformer-based dif fusion model generates layouts, which LLMs refine into hierarchical design s. PCG then constructs high-quality buildings, achieving state-...\n\n\nJu nming Huang, Chi Wang, Letian Li, and Changxin Huang (State Key Laboratory of CAD & CG, Zhejiang University; LIGHTSPEED); Qiang Dai (LIGHTSPEED); an d Weiwei Xu (State Key Lab CAD&CG, Zhejiang University, ZJU-Tencent Game a nd Intelligent Graphics Innovation Technology Joint Lab)\n---------------- -----\nStructRe: Rewriting for Structured Shape Modeling\n\nThe paper pres ents StructRe, a structure rewriting system for 3D shape modeling. It uses an iterative process to rewrite objects, either upwards to more concise s tructures or downwards to more detailed ones, generating hierarchies. This localized rewriting approach enables probabilistic modeling of ...\n\n\nJ iepeng Wang (The University of Hong Kong, Microsoft Research Asia); Hao Pa n (Microsoft Research Asia, Tsinghua University); Yang Liu and Xin Tong (M icrosoft Research Asia); Taku Komura (The University of Hong Kong); Wenpin g Wang (Texas A&M University); and Jiepeng Wang\n---------------------\nIn teractive Optimization of Scaffolded Procedural Patterns\n\nWe introduce a method for interactive design of procedural patterns, allowing users to s ketch content incrementally in a level-by-level fashion. Each level, or sc affold, builds on the previous one, making optimization more responsive an d controllable. A comprehensive validation demonstrates improved...\n\n\nD avide Sforza (Sapienza University of Rome); Marzia Riso (Sapienza Universi ty of Rome; INRIA, Université Côte d'Azur); and Filippo Muzzini, Nicola Ca podieci, and Fabio Pellacini (University of Modena and Reggio Emilia)\n\nI nterest Area: Arts & Design, Gaming & Interactive, New Technologies, Produ ction & Animation, Research & Education\n\nRecording: Livestreamed, Not Li vestreamed, Recorded, Not Recorded\n\nKeyword: Animation, Art, Artificial Intelligence/Machine Learning, Audio, Augmented Reality, Capture/Scanning, Computer Vision, Digital Twins, Display, Dynamics, Education, Ethics and Society, Fabrication, Games, Generative AI, Geometry, Image Processing, Mo deling, Performance, Physical AI, Real-Time, Robotics, Scientific Visualiz ation, Simulation, Spatial Computing, Virtual Reality\n\nRegistration Cate gory: Full Conference, Virtual Access, Experience, Sunday END:VEVENT END:VCALENDAR