BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/Denver X-LIC-LOCATION:America/Denver BEGIN:DAYLIGHT TZOFFSETFROM:-0700 TZOFFSETTO:-0600 TZNAME:MDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0600 TZOFFSETTO:-0700 TZNAME:MST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20250522T212948Z LOCATION:Bluebird Ballroom DTSTART;TZID=America/Denver:20240728T180000 DTEND;TZID=America/Denver:20240728T204500 UID:siggraph_SIGGRAPH 2024_sess211@linklings.com SUMMARY:Papers Fast Forward DESCRIPTION:Sponsored by Adobe: Papers Fast Forward\n\nDeep Hybrid Camera Deblurring for Smartphone Cameras\n\nWe introduce HCDeblur, a practical de blurring framework that utilizes smartphone cameras as a hybrid camera sys tem. Our camera system simultaneously captures a long-exposure image and s hort-exposure burst images. HCDeblur exploits the burst images to effectiv ely remove blur from the long-exposure im...\n\n\nJaesung Rim (POSTECH), J unyong Lee (Samsung AI Center Toronto), and Heemin Yang and Sunghyun Cho ( POSTECH)\n---------------------\nReal-time Hair Rendering With Hair Meshes \n\nWe present how the hair mesh structure can be used for efficiently ren dering strand-based hair models on the GPU. We demonstrate an unprecedente d level of performance for real-time hair rendering, allowing scenes invol ving hundreds of characters with full-resolution unique hair models at rea l-time f...\n\n\nGaurav Bhokare (University of Utah); Eisen Montalvo (Univ ersity of Utah, RTX); Elie Diaz (University of Utah); and Cem Yuksel (Cybe r Radiance)\n---------------------\nNeural Slicer for Multi-axis 3D Printi ng\n\nWe employ neural networks to establish a deformation mapping for cur ved layer generation that can be optimized through loss functions directly defined on the local printing directions. Our method relies less on the i nitial values and can generate results with significantly improved perform ance for m...\n\n\nTao Liu, Tianyu Zhang, Yongxue Chen, Yuming Huang, and Charlie C.L. Wang (University of Manchester)\n---------------------\nDMHom o: Learning Homography with Diffusion Models\n\nDMHomo leverages diffusion -models to generate a realistic dataset for homography learning. We train the generative model using pseudo labels. Additionally, we introduce an it erative process that improves both the homography estimator and the diffus ion models successively, producing a qualified datas...\n\n\nHaipeng Li (U niversity of Electronic Science and Technology of China), Hai Jiang (Sichu an University), Ao Luo (Megvii Technology Limited), Ping Tan (Hong Kong Un iversity of Science and Technology), Haoqiang Fan (Megvii Technology Limit ed), Bing Zeng and Shuaicheng Liu (University of Electronic Science and Te chnology of China), and Haipeng Li\n---------------------\nEulerian-Lagran gian Fluid Simulation on Particle Flow Maps\n\nWe introduce Particle Flow Maps (PFM), a fluid simulation method outperforming Neural Flow Maps (NFM) in efficiency and maintaining high accuracy. By leveraging particle traje ctories and an Eulerian-Lagrangian framework, PFM achieves up to 49 times faster computation and 41\% less memory usage while...\n\n\nJunwei Zhou (U niversity of Michigan), Duowen Chen and Molin Deng (Georgia Institute of T echnology), Yitong Deng (Stanford University), Yuchen Sun (Georgia Institu te of Technology), Sinan Wang (University of Hong Kong), Shiying Xiong (Zh ejiang University), and Bo Zhu (Georgia Institute of Technology)\n-------- -------------\nN-Dimensional Gaussians for Fitting of High Dimensional Fun ctions\n\nWe propose the use of N-Dimensional Gaussian mixtures to fit hig h dimensional functions, applicable to various applications such as neural global illumination and novel view synthesis for complex anisotropic effe cts. Our method offers quality similar to implicit methods with orders of magnitude fast...\n\n\nStavros Diolatzis, Tobias Zirr, and Alexander Kuzne tsov (Intel Labs); Georgios Kopanas (INRIA, Université Côte d'Azur); and A nton Kaplanyan (Intel Labs)\n---------------------\n3D-Layers: Bringing La yer-based Color Editing to VR Painting\n\nWe introduce 3D-Layers, a new da ta structure and companion rendering algorithm to perform\nlayered paintin g in VR. We define 3D-Layers as groups of 3D strokes, and we distinguish t he ones that represent 3D geometry from the ones that represent color modi fications of the geometry.\n\n\nEmilie Yu (Centre Inria d'Université Côte d'Azur), Fanny Chevalier and Karan Singh (University of Toronto), and Adri en Bousseau (Centre Inria d'Université Côte d'Azur)\n--------------------- \nRecompose Grammars for Procedural Architecture\n\nRecomp overcomes funda mental limitations of split grammars by introducing structural shape recom position and geometry component tagging. It allows for selectively reversi ng the subdivision process, enabling a flexible decompose-refine-recompose workflow and precise geometric control, leading to sign...\n\n\nNiklaus H ouska, Cheryl Lau, and Matthias Specht (Esri R&D Center Zurich)\n--------- ------------\nRealFill: Reference-driven Generation for Authentic Image Co mpletion\n\nGiven a few reference images that roughly capture the same sce ne, and a target image with a missing region, RealFill is able to complete the target image with high-quality image content that is faithful to the true scene.\n\n\nLuming Tang (Cornell University); Nataniel Ruiz, Qinghao Chu, Yuanzhen Li, Aleksander Holynski, and David E. Jacobs (Google Researc h); Bharath Hariharan (Cornell University); Yael Pritch and Neal Wadhwa (G oogle Research); Kfir Aberman (Snap); and Michael Rubinstein (Google Resea rch)\n---------------------\nPart123: Part-aware 3D Reconstruction From a Single-view Image\n\nThis paper proposes Part123, a novel framework for pa rt-aware 3D reconstruction from a single-view image. Based on the multivie w images generated by multiview diffusion, we apply contrastive learning t o incorporate their 2D image segmentations into the 3D reconstruction proc ess, thus enabling the ge...\n\n\nAnran Liu, Cheng Lin, Yuan Liu, Xiaoxiao Long, and Zhiyang Dou (University of Hong Kong); Hao-Xiang Guo (Tsinghua University); Ping Luo (University of Hong Kong); and Wenping Wang (Texas A &M University)\n---------------------\nA Heat Method for Generalized Signe d Distance\n\nThe signed distance function (SDF) is a fundamental quantity in geometry processing, but the SDF is not well defined for geometry with topological or geometric errors such as holes, outliers, and self-interse ctions. Our method computes a _generalized signed distance_ directly from broken curves and ...\n\n\nNicole Feng and Keenan Crane (Carnegie Mellon U niversity)\n---------------------\nEfficient Position-based Deformable Col on Modeling for Endoscopic Procedures Simulation\n\nLongstanding challenge s hinder the realism of surgery simulators. Explicitly focusing on colonos copy, traditional algorithms fail to represent the complexities of the end oscope interacting with the colon. Our approach, rooted in XPBD with a Cos serat rod constraint and a novel contact detection algor...\n\n\nMarcelo M artins and Lucas Morais (Federal University of Rio Grande do Sul); Rafael Torchelsen (Federal University of Pelotas); Luciana Nedel (Federal Univers ity of Rio Grande do Sul); and Anderson Maciel (Instituto Superior Técnico , Universidade de Lisboa; Federal University of Rio Grande do Sul)\n------ ---------------\nRadar Fields: Frequency-space Neural Scene Representation s for FMCW Radar\n\nWe present Radar Fields: a frequency-space neural scen e reconstruction method designed for active radar imagers. For the first t ime, we recover dense 3D scene geometry from a single trajectory of 2D rad ar measurements using an explicit mm-wave forward model and an implicit ne ural geometry and reflec...\n\n\nDavid Borts, Erich Liang, and Jipeng Sun (Princeton University); Tim Broedermann, David Brueggemann, and Christos S akaridis (ETH Zürich); Luc Van Gool (ETH Zürich, KU Leuven); and Andrea Ra mazzina, Edoardo Palladin, Stefanie Walz, Mario Bijelic, and Felix Heide ( Princeton University)\n---------------------\nRGB↔X: Image Decomposition a nd Synthesis Using Material- and Lighting-aware Diffusion Models\n\nWe pre sent models for image decomposition into intrinsic channels (RGB→X) and im age synthesis from such channels (X→RGB) in a unified conditional diffusio n framework. We believe it can bring benefits to a wide range of downstrea m editing tasks including material editing, relighting, and ...\n\n\nZheng Zeng (University of California Santa Barbara); Valentin Deschaintre, Iliy an Georgiev, Yannick Hold-Geoffroy, Yiwei Hu, and Fujun Luan (Adobe Resear ch); Ling-Qi Yan (University of California Santa Barbara); and Miloš Hašan (Adobe Research)\n---------------------\nOne Noise to Rule Them All: Lear ning a Unified Model of Spatially-Varying Noise Patterns\n\nWe present a g enerative model that can synthesize a diverse array of spatially-varying n oise patterns, even without spatially-varying training data. Our model off ers a versatile and controllable way to generate noise textures for comput er graphics applications. We also demonstrate its utility in inv...\n\n\nA rman Maesumi, Dylan Hu, and Krishi Saripalli (Brown University); Vladimir Kim and Matthew Fisher (Adobe Research); Soeren Pirk (Kiel University); an d Daniel Ritchie (Brown University)\n---------------------\nA Differential Monte Carlo Solver for the Poisson Equation\n\nWe introduce a general tec hnique that differentiates solutions to the Poisson equation with Dirichle t boundary conditions. Specifically, we devise a new boundary-integral for mulation for the derivatives with respect to arbitrary parameters includin g shapes of the domain. Further, we develop an effic...\n\n\nZihan Yu (Uni versity of California Irvine); Lifan Wu (NVIDIA); Zhiqian Zhou (University of California Irvine); and Shuang Zhao (University of California Irvine, NVIDIA)\n---------------------\nA Dual-Particle Approach for Incompressibl e SPH Fluids\n\nWe propose a dual-particle approach to deal with tensile i nstability in particle methods in fluid simulation, which involves incorpo rating supplementary virtual particles designed to capture and store parti cle pressures. Our approach can accurately simulate free-surface flows wit h rich small-scale t...\n\n\nShusen Liu, Xiaowei He, and Yuzhong Guo (Chin ese Academy of Sciences Institute of Software); Yue Chang (University of Toronto); Wencheng Wang (Chinese Academy of Sciences Institute of Softwar e); and Shusen Liu\n---------------------\nSelf-supervised Video Defocus D eblurring With Atlas Learning\n\nMisfocus is ubiquitous for almost all vid eo producers, degrading video quality and often causing expensive delays a nd reshoots. We address this problem by introducing a self-supervised vide o defocus deblurring pipeline consisting of layered atlases, COC map estim ation, and lens reblur, allowing cons...\n\n\nLingyan Ruan, Mojtaba Bemana , Krzysztof Wolski, Martin Bálint, Hans-Peter Seidel, and Karol Myszkowski (Max Planck Institute for Informatics) and Bin Chen (Max Planck Institute for Informatics, University of Melbourne)\n---------------------\nLightni ng Artist Toolkit: A Hand-Drawn Volumetric Animation Pipeline\n\nWe propos e a set of methods for freely integrating live-action volumetric video wit h frame-by-frame animation created using 6DoF drawing tools—the Lightning Artist Toolkit (Latk), a complete pipeline for hand-drawn volumetric anima tion, and as far as we know the only open-source example of its...\n\n\nNi ck Fox-Gieg (York University)\n---------------------\nAn Induce-on-Boundar y Magnetostatic Solver for Grid-based Ferrofluids\n\nThis paper introduces a novel Induce-on-Boundary (IoB) solver designed to address the magnetost atic problems based on a single-layer potential, which utilizes only the s urface point cloud of the object and offers a lightweight, fast, and accur ate solution for simulating ferrofluids.\n\n\nXingyu Ni (School of CS & Na tional Key Lab of General AI, Peking University); Ruicheng Wang (Peking Un iversity); Bin Wang (Beijing Institute for General Artificial Intelligence ); and Baoquan Chen (School of Intelligence Science and Technology, Peking University)\n---------------------\nSuper-resolution Cloth Animation With Spatial and Temporal Coherence\n\nWe propose a framework for enhancing cl oth animations with high-resolution details while maintaining spatial and temporal consistency. Our method combines a simulator with a GNN corrector to address discrepancies in low-resolution mesh sequences and uses an ima ge-based, super-resolution network to e...\n\n\nJiawang Yu (Style3D Resear ch, Zhejiang University) and Zhendong Wang (Style3D Research)\n----------- ----------\nTEDi: Temporally-Entangled Diffusion for Long-term Motion Synt hesis\n\nTEDi introduces a novel approach to long-term motion synthesis by entangling the denoising diffusion time-axis with the temporal motion tim e-axis. This is achieved through temporally varying denoising, allowing TE Di to generate high-quality, diverse motions auto-regressively. This new m echanism pave...\n\n\nZihan Zhang, Richard Liu, and Rana Hanocka (Universi ty of Chicago) and Kfir Aberman (Snap)\n---------------------\nLearning Ph ysically Realizable Skills for Online Packing of General 3D Shapes\n\nWe s tudy the problem of learning online packing skills for irregular 3D shapes . The goal is to consecutively move a sequence of 3D objects with arbitrar y shapes into a designated container with only partial sequence observatio ns. Our approach considers physical realizability, involving physics dynam ...\n\n\nHang Zhao (School of Computer Science National University of Defe nse Technology), Zherong Pan (Tencent - Lightspeed Studio), Yang Yu (Natio nal Key Laboratory for Novel Software Technology), Kai Xu (School of Compu ter Science National University of Defense Technology), and Hang Zhao\n--- ------------------\nSpice-E: Structural Priors in 3D Diffusion Using Cross -Entity Attention\n\nText-to-3D diffusion models can generate high-quality 3D shapes in seconds, but they are hard to control. In this work we intro duce Spice-E — a neural network that adds structural guidance to 3D diffus ion models, allowing for solving a variety of 3D to 3D tasks with SOTA per formance.\n\n\nEtai Sella, Gal Fiebelman, Noam Atia, and Hadar Averbuch-El or (Tel Aviv University)\n---------------------\nNeural Geometry Fields fo r Meshes\n\nNeural geometry fields are a novel 3D surface representation w hich leverage advances in neural implicit functions to represent discrete triangle meshes. By discretizing a surface into quadrangular patches, a ne ural function can be adapted to learn high frequency surface details, ulti mately enabling ...\n\n\nVenkataram Edavamadathil Sivaram, Tzu-Mao Li, and Ravi Ramamoorthi (University of California San Diego)\n------------------ ---\nLOOSECONTROL: Lifting ControlNet for Generalized Depth Conditioning\n \nLooseControl introduces a generalized approach for depth-conditioned ima ge generation, overcoming ControlNet's reliance on detailed depth maps. It enables scene creation with boundary and 3D box controls for object layou t. This method simplifies complex environment design, showing promise as a versa...\n\n\nShariq Bhat (King Abdullah University of Science and Techno logy (KAUST)); Niloy Mitra (University College London (UCL), Adobe Researc h); and Peter Wonka (King Abdullah University of Science and Technology (K AUST))\n---------------------\nDreamMat: High-quality PBR Material Generat ion With Geometry- and Light-aware Diffusion Models\n\nThis paper introduc es DreamMat, a novel approach for creating high-quality appearances on an untextured mesh by generating PBR materials from a geometry- and light-awa re diffusion model. The resulting PBR materials are free from baked-in lig hting and shadows, suitable for applications in gaming or f...\n\n\nYuqing Zhang (Zhejiang University; State Key Lab CAD&CG, Zhejiang University, ZJ U-Tencent Game and Intelligent Graphics Innovation Technology Joint Lab); Yuan Liu (Tencent Technology); Zhiyu Xie (Zhejiang University; State Key L ab CAD&CG, Zhejiang University, ZJU-Tencent Game and Intelligent Graphics Innovation Technology Joint Lab); Lei Yang (Tencent Technology, Bournemout h University); Zhongyuan Liu, Mengzhou Yang, Runze Zhang, Qilong Kou, and Cheng Lin (Tencent Technology); Wenping Wang (Texas A&M University); and X iaogang Jin (Zhejiang University; State Key Lab of CAD&CG, Zhejiang Univer sity)\n---------------------\nEyeIR: Single Eye Image Inverse Rendering in the Wild\n\nWe propose a method to decompose a single eye image into albe do, shading, specular, normal, and illumination component. We create a div erse synthetic eye dataset and design a synthetic-to-real adaptation frame work for self-supervised learning on real images. We also design a method which specificall...\n\n\nShijun Liang (Beihang University); Haofei Wang ( Peng Cheng Laboratory); and Feng Lu (Beihang University, Peng Cheng Labora tory)\n---------------------\nPreconditioned Nonlinear Conjugate Gradient Method for Real-time Interior-point Hyperelasticity\n\nA preconditioned no nlinear conjugate gradient method is proposed for real-time simulation of elastic deformation with incremental potential contact. This method is GPU -parallelizable and demonstrates fast convergence. A line search strategy is proposed to determine an appropriate step size and achie...\n\n\nXing S hen, Runyuan Cai, Mengxiao Bi, and Tangjie Lv (Fuxi AI Lab, NetEase Inc)\n ---------------------\nCoin3D: Controllable and Interactive 3D Assets Gene ration With Proxy-guided Conditioning\n\nCoin3D is a controllable and inte ractive 3D assets generation framework, which integrates 3D-aware controll ability from coarse geometry to the 3D generation using a volumetric-based 3D adapter. Our interactive modeling workflow facilitates seamless part-a ware editing, allows for previewing the edite...\n\n\nWenqi Dong (Zhejiang University); Bangbang Yang, Lin Ma, and Xiao Liu (ByteDance Inc.); Liyuan Cui and Hujun Bao (Zhejiang University); Yuewen Ma (ByteDance Inc.); and Zhaopeng Cui (Zhejiang University)\n---------------------\nGPT-ME: A Human -AI Cognitive Assemblage\n\nGPT and I are integrated into one another's ex istence through a wearable device. GPT is attached to my body, converting my conversations into prompts. My words reflect GPT's generated outputs ra ther than my own thoughts. I speak GPT. In this symbiotic relationship, my intelligence turns artificial, ...\n\n\nAvital Meshi (University of Calif ornia Davis)\n---------------------\nSemantic Shape Editing With Parametri c Implicit Templates\n\nWe introduce a new way to edit meshes using interp retable parameters by leveraging a parametric implicit representation as t he underlying template. For a generic parametric implicit, we fit the temp late to the input mesh and realize parametric edits on the mesh using our mesh deformation algorithm t...\n\n\nUday Kusupati (EPFL) and Mathieu Gail lard, Jean-Marc Thiery, and Adrien Kaiser (Adobe Research)\n-------------- -------\nPhysics-based Scene Layout Generation From Human Motion\n\nWe pre sent a physics-based framework that generates scene layouts from captured human motions. By simultaneously optimizing the motion imitation character controller and the scene layout generator, our method generates physicall y plausible and semantically reasonable scenes for a range of motions.\n\n \nJianan Li and Tao Huang (The Chinese University of Hong Kong), Qingxu Zh u (Tencent Robotics X), and Tien-Tsin Wong (The Chinese University of Hong Kong)\n---------------------\nBilateral Guided Radiance Field Processing\ n\nGiven a set of multi-view images with photometric variation, our method reconstructs a high-quality radiance field without "floaters" by disentan gling the inconsistent camera processing across different views. Furthermo re, we propose a radiance-finishing approach that can lift user-provided 2 D retou...\n\n\nYuehao Wang, Chaoyi Wang, Bingchen Gong, and Tianfan Xue ( The Chinese University of Hong Kong)\n---------------------\nNeRF as a Non -distant Environment Emitter in Physics-based Inverse Rendering\n\nWe prop ose utilizing NeRF as a non-distant environment lighting model in an inver se rendering pipeline. We demonstrate that our NeRF-based emitter more pre cisely models scene lighting than the conventional environment map, conseq uently enhancing the accuracy of inverse rendering.\n\n\nJingwang Ling, Ru ihan Yu, and Feng Xu (Tsinghua University); Chun Du (Tibet University); an d Shuang Zhao (University of California Irvine)\n---------------------\nRe al-time Wing Deformation Simulations for Flying Insects\n\nThis paper pres ents an efficient skeleton-driven model specifically designed to real-time simulate realistic wing deformations across a wide range of flying insect s. Various simulation experiments, comparisons, and user studies demonstra ted the effectiveness, robustness, and adaptability of the prop...\n\n\nQi ang Chen (Jiangxi University Finance and Economics), Zhigang Deng (Univers ity of Houston), Feng Li and Yuming Fang (Jiangxi University Finance and E conomics), Tingsong Lu and Yang Tong (East China Jiaotong University), and Yifan Zuo (Jiangxi University Finance and Economics)\n------------------- --\nfVDB : A Deep-learning Framework for Sparse, Large Scale, and High Per formance Spatial Intelligence\n\nWe introduce fVDB, a GPU-optimized framew ork for deep learning on large-scale 3D data that efficiently accommodates spatial sparsity, based on a novel VDB index grid structure. Our framewor k is fully integrated with PyTorch and includes a comprehensive collection of operators for tasks such as convo...\n\n\nFrancis Williams, Jiahui Hua ng, Jonathan Swartz, Gergely Klar, Vijay Thakkar, Matthew Cong, and Xuanch i Ren (NVIDIA Research); Ruilong Li (NVIDIA Research, University of Califo rnia Berkeley); Clement Fuji Tsang and Sanja Fidler (NVIDIA Research); Eft ychios Sifakis (University of Wisconsin-Madison, NVIDIA Research); and Ken Museth (NVIDIA Research)\n---------------------\nNeural Bounding\n\nOur r esearch introduces a neural approach to bounding volumes, conservatively c lassifying space across diverse dimensions and scenes. The key is a novel loss function that produces minimal false negatives. Our method extends to non-neural and hybrid representations. We also propose an early exit str. ..\n\n\nStephanie Wenxin Liu (Birkbeck, University of London); Michael Fis cher (University College London (UCL)); Paul D. Yoo (Birkbeck, University of London); and Tobias Ritschel (University College London (UCL))\n------- --------------\nSeparate-and-Enhance: Compositional Finetuning for Text-to -image Diffusion Models\n\nThis work targets on improving the compositiona l capability of text-to-image models. Different from previous approaches t hat requires heavy test-time adaptation per prompt, we propose a compositi onal finetuning framework with two novel objectives. Through comprehensive evaluations, our model demonst...\n\n\nZhipeng Bao (Carnegie Mellon Unive rsity), Yijun Li and Krishna Kumar Singh (Adobe Research), Yu-Xiong Wang ( University of Illinois Urbana-Champaign), and Martial Hebert (Carnegie Mel lon University)\n---------------------\nFrom Microfacets to Participating Media: A Unified Theory of Light Transport With Stochastic Geometry\n\nWe derive a theory of light transport on stochastic implicit surfaces, a geom etry model capable of expressing deterministic geometry, microfacet surfac es, participating media, and an exciting new continuum in between containi ng aggregate appearance, non-classical media, and more. Our model naturall y...\n\n\nDario Seyb (Dartmouth College); Eugene d'Eon and Benedikt Bitter li (NVIDIA); and Wojciech Jarosz (Dartmouth College, NVIDIA)\n------------ ---------\nTexSliders: Diffusion-based Texture Editing in CLIP Space\n\nWe propose a novel, diffusion-based approach for texture editing. We define editing directions using simple text prompts, map these to CLIP image-embe dding space, and project the directions to a CLIP subspace that minimizes identity variations. Our editing pipeline facilitates the creation of arbi tr...\n\n\nJulia Guerrero-Viu (Universidad de Zaragoza), Milos Hasan and A rthur Roullier (Adobe Research), Midhun Harikumar (Adobe), Yiwei Hu and Pa ul Guerrero (Adobe Research), Diego Gutiérrez and Belen Masia (Universidad de Zaragoza), and Valentin Deschaintre (Adobe Research)\n---------------- -----\nInteractive Character Control With Auto-regressive Motion Diffusion Models\n\nWe present A-MDM, an auto-regressive diffusion model for kinema tic motion synthesis. A-MDM can be effectively trained with large motion d atasets to synthesize high-quality and diverse motions. Once trained, A-MD M can be combined with various control methods to generate motions for new downstream tas...\n\n\nYi Shi (Simon Fraser University, Shanghai Aritific ial Intelligence Laboratory); Jingbo Wang and Xuekun Jiang (Shanghai Ariti ficial Intelligence Laboratory); Bingkun Lin (Xmov); Bo Dai (Shanghai Arit ificial Intelligence Laboratory); and Xue Bin Peng (Simon Fraser Universit y, NVIDIA)\n---------------------\nCharacterGen: Efficient 3D Character Ge neration From Single Images With Multi-view Pose Canonicalization\n\nChara cterGen introduces a streamlined 3D character generation pipeline, which c an generate high-quality canonicalized characters from given images within one minute. Additionally, we have curated a dataset, Anime3D, which rende rs anime characters in multiple poses and views to train and evaluate our. ..\n\n\nHao-Yang Peng, Jia-Peng Zhang, and Meng-Hao Guo (Tsinghua Universi ty); Yan-Pei Cao (VAST); and Shi-Min Hu (Tsinghua University)\n----------- ----------\nSpatial and Surface Correspondence Field for Interaction Trans fer\n\nWe introduce a new method for the task of interaction transfer. Giv en an example interaction between a source object and an agent, our method can automatically infer both surface and spatial relationships for the ag ent and target objects within the same category, yielding more accurate an d valid tra...\n\n\nZeyu Huang and Honghao Xu (Shenzhen University), Haibi n Huang and Chongyang Ma (Kuaishou Technology), and Hui Huang and Ruizhen Hu (Shenzhen University)\n---------------------\nReal-time Path Guiding Us ing Bounding Voxel Sampling\n\nWe propose a real-time path guiding method, Voxel Path Guiding (VXPG), that significantly improves fitting efficiency under limited sampling budget. We show that our method can outperform oth er real-time path guiding and virtual point light methods, particularly in handling complex dynamic scenes.\n\n\nHaolin Lu, Wesley Chang, Trevor Hed strom, and Tzu-Mao Li (University of California San Diego)\n-------------- -------\nTowards Unstructured Unlabeled Optical Mocap: A Video Helps!\n\nW e introduce the problem Unstructured Unlabeled Optical (UUO) mocap, where unlabeled optical mocap markers can be placed anywhere on the body. Using monocular video, we introduce a multi-stage optimization framework that le verages multiple hypothesis testing to automatically solve for human pose and...\n\n\nNicholas Milef and John Keyser (Texas A&M University) and Shu Kong (Texas A&M University, University of Macau)\n---------------------\nN eural Monte Carlo Fluid Simulation\n\nWe present a novel neural network re presentation for fluid simulation that augments neural fields with explici tly enforced boundary conditions and a Monte Carlo pressure solver to elim inate all weakly enforced boundary conditions. Our method is mesh-free and can accurately represent vorticity phenom...\n\n\nPranav Jain (University of Southern California); Ziyin Qu (University of Pennsylvania); Peter Yic hen Chen (Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology (MIT)); and Oded Stein (University of Southern California)\n---------------------\nFactorized Motion Fields f or Fast Sparse Input Dynamic View Synthesis\n\nThe performance of dynamic radiance fields reduces significantly with sparse input viewpoints. We des ign a fast and explicit motion-model using factorized volumes and regulari ze with flow priors. Since cross-camera dense-flow-priors are unreliable, we obtain reliable flow priors as a combination of ...\n\n\nNagabhushan So mraj, Kapil Choudhary, Sai Harsha Mupparaju, and Rajiv Soundararajan (Indi an Institute of Science)\n---------------------\nSubject-Diffusion: Open D omain Personalized Text-to-image Generation Without Test-time Fine-tuning\ n\nWe present Subject-Diffusion, a novel open-domain personalized image ge neration model that, in addition to not requiring test-time fine-tuning, a lso only requires a single reference image to support personalized generat ion of single- or two-subjects in any domain.\n\n\nJian Ma (OPPO), Junhao Liang (Southern University of Science and Technology), and Chen Chen and H aonan Lu (OPPO)\n---------------------\nVersatile Vision Foundation Model for Image and Video Colorization\n\nIn this work, we show how a latent dif fusion model, pre-trained on text-to-image synthesis, can be repurposed fo r image colorization and provide a flexible high-quality solution for a wi de variety of scenarios: direct colorization with diverse results, user gu idance through colors hints or text prom...\n\n\nVukasin Bozic (ETH Zürich ); Abdelaziz Djelouah and Yang Zhang (Disney Research Studios); Radu Timof te (University of Wurzburg); Markus Gross (ETH Zürich, Disney Research Stu dios); and Christopher Schroers (Disney Research Studios)\n--------------- ------\nReCollection: Creating synthetic memories with AI in an interactiv e art installation\n\nThis work integrates AI system design with experimen tal data visualization, providing a collaborative AI art experience that i s intimate, accessible, interactive, unpredictable, and immersive. Beyond its potential as a future therapeutic prototype for dementia groups, this work questions and reflect...\n\n\nWeidi Zhang (Arizona State University), Lijiaozi Cheng (University of Sheffield), and Jieliang Luo (Autodesk Rese arch)\n---------------------\nPhysics-informed Learning of Characteristic Trajectories for Smoke Reconstruction\n\nWe introduce Neural Characteristi c Trajectory Fields, a novel representation utilizing Eulerian neural fiel ds to implicitly model Lagrangian fluid trajectories for video-based fluid reconstruction. This topology-free, auto-differentiable representation fa cilitates end-to-end supervision, encompassin...\n\n\nYiming Wang and Siyu Tang (ETH Zürich) and Mengyu Chu (Peking University, National Key Lab of General AI)\n---------------------\nDAE-Net: Deforming Auto-Encoder for Fi ne-grained Shape Co-segmentation\n\nWe present an unsupervised 3D shape co -segmentation method following the stipulation that corresponding parts in different shapes should have approximately the same shape. Our method lea rns the shapes of a set of part templates and composes each shape by selec ting a subset of template parts which ar...\n\n\nZhiqin Chen (Adobe Resear ch) and Qimin Chen, Hang Zhou, and Hao Zhang (Simon Fraser University)\n-- -------------------\nColorVideoVDP: A Visual Difference Predictor for Imag e, Video, and Display Distortions\n\nColorVideoVDP is a differentiable ima ge and video quality metric that models human color and spatiotemporal vis ion. It is targeted and calibrated to assess image distortions due to AR/V R display technologies and video streaming, and it can handle both SDR and HDR content.\n\n\nRafal K. Mantiuk, Param Hanji, and Maliha Ashraf (Unive rsity of Cambridge) and Yuta Asano and Alexandre Chapiro (Meta)\n--------- ------------\nStreetscapes: Large-scale Consistent Street View Generation Using Autoregressive Video Diffusion\n\nOur method generates Streetscapes — long sequences of views through a synthesized city-scale scene. We build on video diffusion models, but in an autoregressive framework that easily scales to long camera trajectories. We train our system on the unique Goo gle Street View data, allowing control...\n\n\nBoyang Deng (Stanford Unive rsity, Google Research); Richard Tucker and Zhengqi Li (Google Research); Leonidas Guibas (Stanford University, Google Research); Noah Snavely (Goog le Research, Cornell University); and Gordon Wetzstein (Stanford Universit y)\n---------------------\nS3: Speech, Script, and Scene Driven Head and E ye Animation\n\nOur method generates 3D head and eye animations for charac ters in conversation using audio, scripts, and scene inputs. It incorporat es animator goals and psycho-linguistic insights, producing realistic anim ations that align with human behavior and compares favorably with existing approaches, as conf...\n\n\nYifang Pan (University of Toronto; Jali Resea rch, Canada); Rishabh Agrawal (Jali Research, Canada); and Karan Singh (Un iversity of Toronto; Jali Research, Canada)\n---------------------\nReal-t ime Neural Woven Fabric Rendering\n\nWe propose a lightweight neural netwo rk to represent woven fabrics at different scales in real time. By encodin g the regular and repetitive woven fabric pattern into a small latent vect or, our network is able to handle three typical types of woven fabrics in our training dataset: plain, twill, and s...\n\n\nXiang Chen and Lu Wang ( Shandong University) and Beibei Wang (Nanjing University)\n--------------- ------\nTaming Diffusion Probabilistic Models for Character Control\n\nWe present a novel character control framework that enables real-time generat ion of high-quality, diverse character animations in response to user inte ractive control. Our framework supports the animation of characters in var ious styles using a single, unified model.\n\n\nRui Chen (Hong Kong Univer sity of Science and Technology), Mingyi Shi (University of Hong Kong), Sha oli Huang (Tencent AI Lab), Ping Tan (Hong Kong University of Science and Technology), Taku Komura (University of Hong Kong), and Xuelin Chen (Tence nt AI Lab)\n---------------------\nLearning a Generalized Physical Face Mo del From Data\n\nIn this work, we aim to democratize physics-based facial animation by proposing a generalized physical face model that we learn fro m a large 3D face dataset. Once trained, our model can be quickly fit to a ny unseen identity and automatically produce a ready-to-animate physical f ace model.\n\n\nLingchen Yang (ETH Zürich); Gaspard Zoss and Prashanth Cha ndran (Walt Disney Company, Switzerland); Markus Gross (ETH Zürich; Walt D isney Company, Switzerland); Barbara Solenthaler (ETH Zürich); Eftychios S ifakis (University of Wisconsin Madison); and Derek Bradley (Walt Disney C ompany, Switzerland)\n---------------------\nMatting by Generation\n\nThis paper redefines traditional regression-based matting as a generative mode ling challenge and harnesses the capabilities of latent diffusion models e nriched with extensive pre-trained knowledge to tackle this challenge. It not only produces mattes with superior resolution and detail but is also v ...\n\n\nZhixiang Wang (University of Tokyo), Baiang Li (Hefei University of Technology), Jian Wang (Snap), Yu-Lun Liu (National Chiao Tung Universi ty), Jinwei Gu (Chinese University of Hong Kong), Yung-Yu Chuang (National Taiwan University), and Shin'ichi Satoh (National Institute of Informatic s)\n---------------------\nI2V-Adapter: A General Image-to-Video Adapter f or Diffusion Models\n\nWe propose I2V-Adapter, a lightweight adapter modul e for Image-to-Video generation. Our method achieves excellent performance in general I2V scenarios and is also compatible with various plugins from the Stable Diffusion open-source community. Additionally, we propose Fram e-Similarity-Prior to contr...\n\n\nXun Guo (University of Science and Tec hnology of China); Mingwu Zheng, Liang Hou, Yuan Gao, Yufan Deng, Pengfei Wan, and Di Zhang (Kuaishou Technology); Yufan Liu and Weiming Hu (Institu te of Automation, Chinese Academy of Sciences); Zhengjun Zha (University o f Science and Technology of China); and Haibin Huang and Chongyang Ma (Kua ishou Technology)\n---------------------\nEASI-Tex: Edge-Aware Mesh Textur ing from Single Image\n\nWe introduce a novel approach for single-image me sh texturing, which employs a pre-trained image diffusion model with judic ious conditioning to seamlessly transfer texture from a single image to a given 3D mesh object, without any optimization or training.\n\n\nSai Raj K ishore Perla, Yizhi Wang, and Ali Mahdavi-Amiri (Simon Fraser University) and Hao Zhang (Simon Fraser University, Amazon)\n---------------------\nSu rface-filling Curve Flows via Implicit Medial Axes\n\nWe introduce a fast, robust, and user-controllable algorithm to generate surface-filling curve s. We compute these curves through the gradient flow of a simple sparse en ergy. Our algorithm makes minimal assumptions on the input surface, achiev ing improved robustness. Our framework provides tuneable p...\n\n\nYuta No ma (University of Tokyo, University of Toronto); Silvia Sellán (University of Toronto); Nicholas Sharp (NVIDIA); Karan Singh (University of Toronto) ; and Alec Jacobson (University of Toronto, Adobe Research)\n------------- --------\nTensoSDF: Roughness-aware Tensorial Representation for Robust Ge ometry and Material Reconstruction\n\nWe propose a novel framework for rob ust geometry and material reconstruction. The framework's core is the roug hness-aware incorporation of the radiance and reflectance fields to recons truct arbitrary reflective objects. The proposed TensoSDF representation e nhances the geometry details while acceler...\n\n\nJia Li (Shandong Univer sity), Beibei Wang (Nanjing University), Lu Wang (Shandong University), an d Lei Zhang (The Hong Kong Polytechnic University)\n---------------------\ nOnline Neural Path Guiding with Normalized Anisotropic Spherical Gaussian s\n\nWe propose a online framework to learn the spatial-varying distributi on of the full product of the rendering equation, with a single small neur al network using stochastic ray samples, and a novel, expressive, closed-f orm density model called the Normalized Anisotropic Spherical Gaussian mix ture.\n\n\nJiawei Huang (Chuzhou University, Void Dimensions); Akito Iizuk a and Hajime Tanaka (Tohoku University); Taku Komura (University of Hong K ong); Yoshifumi Kitamura (Tohoku University); and Jiawei Huang\n---------- -----------\nObject-level Scene Deocclusion\n\nIn this paper, we present a new self-supervised framework, named PACO, for object-level scene deocclu sion to deocclude each of the objects of a real-world scene. Our approach allows multiple downstream applications, including scene-level, single-im age 3D reconstruction and object rearrangement in i...\n\n\nZhengzhe Liu ( The Chinese University of Hong Kong); Qing Liu (Adobe Research); Chirui Ch ang (University of Hong Kong); Jianming Zhang, Daniil Pakhomov, Haitian Zh eng, and Zhe Lin (Adobe Research); Daniel Cohen-Or (Tel Aviv University); and Chi-Wing Fu (The Chinese University of Hong Kong)\n------------------- --\nFilter-Guided Diffusion for Controllable Image Generation\n\nFilter-Gu ided Diffusion (FGD) is a controllable, tuning-free, image-to-image transl ation method for diffusion models. It combines fast filtering operations w ith non-deterministic samplers to generate high-quality and diverse images . With its efficiency, FGD can be sampled multiple times to outperfor...\n \n\nZeqi Gu (Cornell-Tech Cornell University) and Ethan Yang and Abe Davis (Cornell University)\n---------------------\nMinkowski Penalties: Robust Differentiable Constraint Enforcement for Vector Graphics\n\nWe introduce a method for laying out complex arrangements of general, nonconvex 2D shap es in the context of vector graphics, illustration, and diagrams. Beyond s imple "no-overlap" conditions, we introduce differentiable penalties for n ested containment, tangency, precise padding, etc., which are rob...\n\n\n Jiří Minarčík (Independent) and Sam Estep, Wode Ni, and Keenan Crane (Carn egie Mellon University)\n---------------------\nMotionCtrl: A Unified and Flexible Motion Controller for Video Generation\n\nMotionCtrl, through the delicate design of architecture, training strategy, and curated datasets, enables independent or combined control of camera and object motion withi n a unified video generation model. Its generalizability extends to existi ng video generation models like LVDM/VideoCrafter1 and ...\n\n\nZhouxia Wa ng (University of Hong Kong); Ziyang Yuan (Tsinghua University); Xintao Wa ng (ARC Lab, Tencent PCG); Yaowei Li (Peking University); Tianshui Chen (G uangdong University of Technology); Menghan Xia (Tencent AI Lab); Ping Luo (University of Hong Kong); and Ying Shan (ARC Lab, Tencent PCG)\n-------- -------------\nA Realistic Multi-scale Surface-based Cloth Appearance Mode l\n\nWe present a multi-scale, surface-based cloth appearance model which is efficient and photo-realistic. We propose a comprehensive micro-scale m odel focusing on correct parallax effects and a practical meso-scale integ ration scheme, emphasizing efficiency while losslessly preserving accurate highligh...\n\n\nJunqiu Zhu (University of California Santa Barbara); Chr istophe Hery, Lukas Bode, Carlos Aliaga, and Adrian Jarabo (Meta Reality L abs Research); Ling-Qi Yan (University of California Santa Barbara); and M att Jen-Yuan Chiang (Meta Reality Labs Research)\n---------------------\nI nteractive Design of Stylized Walking Gaits for Robotic Characters\n\nWe p resent a real-time procedural animation technique for the artist-directed authoring of stylized walking gaits for robotic characters. Our interactiv e authoring tool relies on a parameteric gait planner and model-based cont rol stack to generate feasible walking motion from example walk cycles usi ...\n\n\nMichael A. Hopkins and Georg Wiedebach (Disney Research); Kyle Ce sare and Jared Bishop (Walt Disney Imagineering R&D, PickNik Robotics); an d Espen Knoop and Moritz Bächer (Disney Research)\n---------------------\n 3Doodle: Compact Abstraction of Objects With 3D Strokes\n\nWe propose 3Doo dle, a compact and efficient representation to convey characteristics of a n object. Our approach generates 3D strokes from multi-view images. We exp ress 3D sketch with contour of superquadrics (view-dependent component) an d 3D cubic Bezier curves (view-independent component). 3Doodle ...\n\n\nCh angwoon Choi, Jaeah Lee, Jaesik Park, and Young Min Kim (Seoul National Un iversity)\n---------------------\nText-to-vector Generation With Neural Pa th Representation\n\nWe propose a novel pipeline to generate high-quality vector graphics based on text prompts. Utilizing neural path representatio n and a two-stage path optimization process, we can incorporate geometric constraints while preserving expressivity in the generated SVGs.\n\n\nPeiy ing Zhang (City University of Hong Kong), Nanxuan Zhao (Adobe Research), a nd Jing Liao (City University of Hong Kong)\n---------------------\nAccele rating Saccadic Response Through Spatial and Temporal Cross-modal Misalign ments\n\nWe investigate how auditory cues affect saccadic latency to a vis ual target, showing that preceding sounds can significantly reduce latency . We study several factors: eccentricity of the visual stimulus and spatio -temporal shifts of the audio stimulus. We discuss various applications an d validate ou...\n\n\nDaniel Jiménez Navarro (Max Planck Institute for Inf ormatics), Xi Peng and Yunxiang Zhang (New York University), Karol Myszkow ski and Hans-Peter Seidel (Max Planck Institute for Informatics), Qi Sun ( New York University), and Ana Serrano (Universidad de Zaragoza)\n--------- ------------\nHQ3DAvatar: High Quality Implicit 3D Head Avatar\n\nWe prese nt a novel method for rendering photorealistic human head avatars. Our met hod utilizes an implicitly learned canonical space constrained using optic al flow in a multiresolution hash encoding framework. Our approach outperf orms related methods and excels in reconstructing regions exhibiting c...\ n\n\nKartik Teotia and Byrasandra Ramalinga Reddy Mallikarjun (Max Planck Institute for Informatics); Xingang Pan (Max Planck Institute for Informat ics, Nanyang Technological University); Hyeongwoo Kim (Imperial College Lo ndon); Pablo Garrido (Flawless AI); Mohamed Elgharib and Christian Theobal t (Max Planck Institute for Informatics); and Kartik Teotia\n------------- --------\nDeadWood: Including disturbance and decay in the depiction of di gital nature\n\nWe present a framework that combines an ecosystem simulati on, which explicitly incorporates disturbance events and decay processes, with a model realization process, which balances the uniqueness arising fr om life history with the need for instancing due to memory constraints.\n\ n\nAdrien Peytavie (Université Lyon 1, CNRS, LIRIS); James Gain (Universit y of Cape Town); Eric Guérin (INSA, Lyon); Oscar Argudo (Universitat Polit ècnica de Catalunya); Eric Galin (Université Lyon 1, CNRS, LIRIS); and Adr ien Peytavie\n---------------------\nCurvature-driven Conformal Deformatio ns\n\nWe introduce a novel approach for computing conformal deformations i n Euclidean space while minimizing curvature-based energies. These energie s serve as fundamental tools in geometry processing, essential for tasks s uch as surface fairing, deformation, approximations using cone metric surf aces. We e...\n\n\nEtienne Corman (CNRS, Inria, LORIA)\n------------------ ---\nPortrait3D: Text-guided High-quality 3D Portrait Generation Using Pyr amid Representation and GANs Prior\n\nWe present Portrait3D, a novel text- to-3D-portrait generation framework that produces high-quality, view-consi stent, realistic, and canonical 3D portraits that are in alignment with th e input text prompts.\n\n\nYiqian Wu, Hao Xu, and Xiangjun Tang (Zhejiang University; State Key Lab of CAD&CG, Zhejiang University); Xien Chen (Yale University); Siyu Tang (ETH Zürich); Zhebin Zhang and Chen Li (OPPO US Re search Center); and Xiaogang Jin (Zhejiang University; State Key Lab of CA D&CG, Zhejiang University)\n---------------------\nInvertAvatar: Increment al GAN Inversion for Generalized Head Avatars\n\nOur "Incremental 3D GAN I nversion" framework advances avatar reconstruction with flexible-number mu ltiple frames, incorporating a unique animatable GAN prior for precise exp ressions control, UV parameterization for high-fidelity appearance recover y, and a ConvGRU-based recurrent network for temporal...\n\n\nXiaochen Zha o, Jingxiang Sun, Lizhen Wang, Jinli Suo, and Yebin Liu (Tsinghua Universi ty)\n---------------------\nSmooth Bijective Projection in a High-order Sh ell\n\nWe propose a new high-order shell structure for the smooth attribut e transfer between meshes inside the shell, along with a robust constructi on algorithm. The high-order shell is enveloped by three B\'{e}zier triang les and three side surfaces, with a smooth bijective projection inside.\n\ n\nShibo Liu, Yang Ji, Jia-Peng Guo, Ligang Liu, and Xiao-Ming Fu (Univers ity of Science and Technology of China)\n---------------------\nFloating o n the Boundary: Perceptions of Reality in a Half-Digital, Half-Physical Bu nny\n\nWe present ‘Float’ as a integrated display method with grasses-free 3D display, and the findings of user study. This research has uncovered t he ‘Physical-Digital Continuous Illusion’—a phenomenon a sense of object e xisting across different spaces. The exploration within...\n\n\nTakumi Yok oyama, Kazuya Izumi, Tatsuki Fushimi, and Yoichi Ochiai (University of Tsu kuba)\n---------------------\nDeep Fourier-based Arbitrary-scale Super-res olution for Real-time Rendering\n\nBy observing that high-resolution G-buf fers possess similar spectrum to high-resolution rendered frame, we propos e a method by supporting arbitrary-scale super-resolution for a trained ne ural model. The key is a Fourier-based implicit neural representation whic h maps arbitrary and naturally continuo...\n\n\nHaonan Zhang, Jie Guo, Jia wei Zhang, Haoyu Qin, Zesen Feng, Ming Yang, and Yanwe Guo (Nanjing Univer sity)\n---------------------\nHandcrafting in Zero Gravity: Reinventing th e Spindle as an Artistic Intervention in Space Research\n\nAchieving diver sity in techno-scientific research is a major theme in today's discourses. However, earlier tendencies continue in practice. The inclusion of people from diverse backgrounds is increasing, but their situated knowledges are not called upon in defining research agendas. The art project ...\n\n\nEb ru Kurbak (University of Applied Arts Vienna)\n---------------------\nCycl ogenesis: Simulating Hurricanes and Tornadoes\n\nWe propose a physically b ased approach to describe the 3D development of cyclones in a visually con vincing and physically plausible manner. Modeling these processes enables us to simulate multiple hurricane and tornado phenomena.\n\n\nJorge Alejan dro Amador Herrera, Jonathan Klein, and Daoming Liu (King Abdullah Univers ity of Science and Technology (KAUST)); Wojtek Pałubicki (AMU); Sören Pirk (CAU); and Dominik L. Michels (King Abdullah University of Science and Te chnology (KAUST))\n---------------------\nNeurCADRecon: Neural Representat ion for Reconstructing CAD Surfaces by Enforcing Zero Gaussian Curvature\n \nIn this paper, we present a self-supervised neural representation aimed at reconstructing CAD models from low-quality, unoriented point clouds. Ou r approach stands out from traditional reconstruction methods by applying a zero Gaussian curvature constraint, which emphasizes the characteristic of bei...\n\n\nQiujie Dong and Rui Xu (Shandong University); Pengfei Wang (University of Hong Kong); Shuangmin Chen (Qingdao University of Science a nd Technology); Shiqing Xin (Shandong University); Xiaohong Jia (AMSS, Chi nese Academy of Sciences); Wenping Wang (Texas A&M University); and Changh e Tu (Shandong University)\n---------------------\nDressCode: Autoregressi vely Sewing and Generating Garments From Text Guidance\n\nDressCode presen ts an advanced Generative AI framework specifically designed for 3D garmen ts. Leveraging the power of natural language, Dresscode incorporates Sewin gGPT for sewing pattern generation and a fine-tuned diffusion model for PB R texture synthesis, which showcases interaction-friendly appl...\n\n\nKai He (ShanghaiTech University, Deemos Technology); Kaixin Yao (ShanghaiTech University, NeuDim); Qixuan Zhang (ShanghaiTech University, Deemos Techno logy); Jingyi Yu (ShanghaiTech University); Lingjie Liu (University of Pen nsylvania); and Lan Xu (ShanghaiTech University)\n---------------------\nW alkin’ Robin: Walk on Stars With Robin Boundary Conditions\n\nWe develop a grid-free Monte Carlo method for solving boundary value problems like the Poisson equation with Dirichlet, Neumann, and Robin boundary conditions. Unlike conventional PDE solvers, our method does not require volumetric me shing or global solves. It is robust, embarrassingly parallel, sca...\n\n\ nBailey Miller (Carnegie Mellon University), Rohan Sawhney (NVIDIA), and K eenan Crane and Ioannis Gkioulekas (Carnegie Mellon University)\n--------- ------------\nModelling a Feather as a Strongly Anisotropic Elastic Shell\ n\nWe investigate the mechanical properties of bird feathers. Our lab expe riments reveal a linear strain-stress relationship of the feather membrane in addition to an extreme anisotropy. From these findings we build a simp le orthotropic model for the feather vane, whose numerical implementation avoids ...\n\n\nJean Jouve and Victor Romero (University Grenoble Alpes In ria, CNRS, Grenoble INP, LJK); Rahul Narain (IIT Delhi); Laurence Boissieu x (INRIA); Theodore Kim (Yale University); and Florence Bertails-Descoubes (University Grenoble Alpes Inria, CNRS, Grenoble INP, LJK)\n------------- --------\nMulti-material Mesh-based Surface Tracking With Implicit Topolog y Changes\n\nWe introduce a multi-material, non-manifold mesh-based surfac e tracking algorithm that converts self-intersections into topological cha nges. Our algorithm preserves surface features like mesh-based methods and robustly handles topological changes like level set methods. We demonstra te the effectiven...\n\n\nPeter Heiss-Synak, Aleksei Kalinov, Malina Strug aru, and Arian Etemadi (Institute of Science and Technology Austria (ISTA) ); Huidong Yang (University of Vienna); and Chris Wojtan (Institute of Sci ence and Technology Austria (ISTA))\n---------------------\nA Concert in a Vanished Church: Contextualizing Peace Island's Auditory History with Mod ern Technology\n\nOur project aims to revitalize archaeological sites thro ugh music technology and transform unearthed artifacts and echoes of past lives into auditory experiences. "Ashes to Ashes," a live performance, and an accompanying documentary translate archaeological findings into innova tive soundscapes using...\n\n\nTak Cheung Hui (Hong Kong Metropolitan Univ ersity) and Yu Chia Kuo (Institute of Information Science, Academia Sinica , Taiwan)\n---------------------\nA Unified Differentiable Boolean Operato r With Fuzzy Logic\n\nWe present a differentiable boolean operator for imp licit solid shape modeling. Drawing inspiration from fuzzy logic, our bool ean operator outputs a continuous function and is differentiable with resp ect to operator types. This enables optimization of the primitives and the boolean operations employ...\n\n\nHsueh-Ti Derek Liu (Roblox, University of British Columbia); Maneesh Agrawala (Stanford University, Roblox); Cem Yuksel (University of Utah, Roblox); Tim Omernick, Vinith Misra, and Stefa no Corazza (Roblox); Morgan McGuire (Roblox, McGill University); and Victo r Zordan (Roblox)\n---------------------\nModeling Ambient Scene Dynamics for Free-view Synthesis\n\nWe introduce an innovative method for view synt hesis of dynamic scenes from one monocular video, enhancing immersion by o vercoming previous limitations of 3D Gaussian Splatting. By exploiting amb ient motion periodicity and regularization, our approach reconstructs deta iled natural scenes with motion...\n\n\nMeng-Li Shih (University of Washin gton); Jia-Bin Huang (University of Maryland, College Park; Meta); and Cha ngil Kim, Rajvi Shah, Johannes Kopf, and Chen Gao (Meta)\n---------------- -----\nInteractive Invigoration: Volumetric Modeling of Trees With Strands \n\nWe introduce an interactive tree modeling approach that mimics tree se condary growth with strand-based volumetric representation. Our model enab les interactive generation of complex detailed tree models with minimal ef fort by integrating strands with an interactive tree development model and a set o...\n\n\nBosheng Li (Purdue University), Nikolas Schwarz (Kiel Uni versity), Wojtek Palubicki (AMU), Soeren Pirk (Kiel University), and Bedri ch Benes (Purdue University)\n---------------------\nReFiNe: Recursive Fie ld Networks for Cross-modal Multi-scene Representation\n\nThe common trade -offs of state-of-the-art methods for multi-shape representation involve t rading modeling accuracy against memory and storage. We show how a recursi ve hierarchical architecture can be used to encode multiple shapes represe nted as continuous neural fields with a higher degree of preci...\n\n\nSer gey Zakharov, Katherine Liu, Adrien Gaidon, and Rares Ambrus (Toyota Resea rch Institute)\n---------------------\nA Free-space Diffraction BSDF\n\nFr ee-space diffractions are an optical phenomenon where light diffracts (“be nds”) around the geometric edges and corners of scene objects. In this pap er, we present an efficient method to simulate such effects and show that we can run wave simulations on scenes orders-of-magnitude more c...\n\n\nS hlomi Steinberg (University of Waterloo); Ravi Ramamoorthi (University of California San Diego, NVIDIA); Benedikt Bitterli (NVIDIA); Arshiya Mollaza inali (University of Waterloo); and Eugene d'Eon and Matt Pharr (NVIDIA)\n ---------------------\nDifferentiable Voronoi Diagrams for Simulation of C ell-based Mechanical Systems\n\nWe present a computational model leveragin g analytical derivatives of Voronoi diagram geometry for simulation of cel l-based mechanical systems. Boundary coupling formulation enables interact ion with rigid bodies, elastic membranes, and free surfaces. Examples mode l complex dynamic processes involvin...\n\n\nLogan Numerow, Yue Li, Stelia n Coros, and Bernhard Thomaszewski (ETH Zürich)\n---------------------\nVa riational Feature Extraction in Scientific Visualization\n\nFeature extrac tion is a common approach to analyze large scientific data sets. At presen t, many extraction techniques exist in different application domains. Usin g variational calculus, we phrase common feature definitions in a consiste nt mathematical language. We apply our framework in fluid dynami...\n\n\nN ico Daßler and Tobias Günther (University of Erlangen-Nuremberg)\n-------- -------------\nNeural Gaussian Scale-space Fields\n\nWe present a method t o learn a fully continuous Gaussian scale-space from raw data. This allows efficient and flexible anisotropic filtering and can be used to create mu ltiscale representations across a broad range of modalities and applicatio ns.\n\n\nFelix Mujkanovic, Ntumba Elie Nsampi, Christian Theobalt, Hans-Pe ter Seidel, and Thomas Leimkühler (Max Planck Institute for Informatics)\n ---------------------\nDiffusion Texture Painting\n\nWe present a techniqu e that leverages 2D generative diffusion models for interactive texture pa inting on the surface of 3D meshes. Unlike existing texture painting syste ms, our method allows artists to paint with any complex image texture. Our brush generates seamless strokes in real-time and inpain...\n\n\nAnita Hu (NVIDIA); Nishkrit Desai (University of Toronto, Vector Institute); Hassa n Abu Alhaija (NVIDIA); Seung Wook Kim (NVIDIA, University of Toronto); an d Maria Shugrina (NVIDIA)\n---------------------\nBlue Noise for Diffusion Models\n\nMost existing diffusion models use Gaussian noise for training. We introduce a novel class of deterministic diffusion models using time-v arying noise (i.e., from white to blue noise) to incorporate correlation w ithin images during training. Further, our framework allows introducing co rrelation acros...\n\n\nXingchang Huang and Corentin Salaun (Max Planck In stitute for Informatics); Cristina Vasconcelos (Google DeepMind); Christia n Theobalt (Max Planck Institute for Informatics); Cengiz Oztireli (Google Research, University of Cambridge); and Gurprit Singh (Max Planck Institu te for Informatics)\n---------------------\nUnveiling New Artistic Dimensi ons in Calligraphic Arabic Script with Generative Adversarial Networks\n\n Our project merges calligraphic Arabic script with GAN-based generative AI , emphasizing the cultural and artistic significance of blending tradition al art with technology. We spotlight generative AI's potential to reimagin e underrepresented art forms and underscore the importance of specialized arti...\n\n\nArshia Sobhan Sarbandi and Philippe Pasquier (Simon Fraser Un iversity) and Adam Tindale (OCAD University)\n---------------------\nMesh Neural Cellular Automata\n\nMeshNCA directly creates dynamic textures on 3 D meshes without UV maps. MeshNCA's training targets include PBR textures, text prompts, and motion-vector fields. Trained only on an Icosphere mesh , we create textures on unseen meshes and interactively edit the synthesiz ed textures, both in real time. ...\n\n\nEhsan Pajouheshgar and Yitao Xu ( EPFL), Alexander Mordvintsev and Eyvind Niklasson (Google Research), and T ong Zhang and Sabine Süsstrunk (EPFL)\n---------------------\nA Hierarchic al 3D Gaussian Representation for Real-time Rendering of Very Large Scenes \n\nWe introduce a hierarchy for 3D Gaussian splatting, merging primitives in a way that preserves speed and quality. We divide the scene into chunk s, each having a hierarchy that is further optimized. A consolidated hiera rchy allows reconstruction and real-time rendering of very large scenes of tens of...\n\n\nBernhard Kerbl (INRIA, Université Côte d'Azur; Technical University of Vienna); Andreas Meuleman and Georgios Kopanas (INRIA, Unive rsité Côte d'Azur); Michael Wimmer (Technische Universität Wien (TU Wien)) ; and Alexandre Lanvin and George Drettakis (Institut national de recherch e en informatique et en automatique (INRIA), Unversité Côte d'Azur)\n----- ----------------\nTemporal Acoustic Point Holography\n\nIn this work, we d evelop temporal phase retrieval algorithms for acoustic point holography. We satisfy acoustic field amplitude objectives, while suppressing transduc ers’ phase change over time, to enable dynamic 3D particle displays with m inimum geometric restrictions. We experimentally demon...\n\n\nGiorgos Chr istopoulos, Lei Gao, Diego Martinez Plasencia, Marta Betcke, Ryuji Hirayam a, and Sriram Subramanian (University College London (UCL))\n------------- --------\nUltra Inertial Poser: Scalable Motion Capture and Tracking From Sparse Inertial Sensors and Ultra-Wideband Ranging\n\nUltra Inertial Poser addresses sparse inertial pose estimation challenges by integrating inter -sensor distances measured through Ultra-Wideband ranging. Its pipeline le verages commodity IMU and UWB sensors, and a graph-based method to integra te distance constraints into existing frameworks. It is su...\n\n\nRayan A rmani, Changlin Qian, Jiaxi Jiang, and Christian Holz (ETH Zürich)\n------ ---------------\nContourCraft: Learning to Resolve Intersections in Neural Multi-garment Simulations\n\nWe present a learning-based solution for han dling collisions and self-intersections. Our method robustly recovers from intersections introduced through missed collisions, self-penetrating bodi es, or manually designed multi-layer outfits. The technical core of our me thod is a novel intersection conto...\n\n\nArtur Grigorev (ETH Zürich, Max Planck Institue for Intelligent Systems); Giorgio Becherini and Michael B lack (Max Planck Institue for Intelligent Systems); and Otmar Hilliges and Bernhard Thomaszewski (ETH Zürich)\n---------------------\nSMEAR: Stylize d Motion Exaggeration With ARt-direction\n\nSmear frames are routinely use d by artists for the expressive depiction of motion in animations. In this paper, we present an automatic, art-directable method and its Blender imp lementation for the generation of smear frames in 3D, with a focus on elon gated in-betweens where an object is stretched a...\n\n\nJean Basset (INRI A, Bordeaux Sud-Ouest); Pierre Bénard (Laboratoire Bordelais de Recherche en Informatique (LaBRI); INRIA, Bordeaux Sud-Ouest); and Pascal Barla (INR IA, Bordeaux Sud-Ouest)\n---------------------\nTransparent Image Layer Di ffusion Using Latent Transparency\n\nWe present an approach enabling large -scale pretrained latent diffusion models like Stable Diffusion to generat e transparent images and layers.\n\n\nLvmin Zhang and Maneesh Agrawala (St anford University)\n---------------------\nCategorical Codebook Matching f or Embodied Character Controllers\n\nThis work presents a generative VQ fr amework to translate movements of a real user onto a virtual embodied avat ar using VR inputs. Our proposed codebook matching technique enables simul taneously learning and sampling the motions in form of Categorical probabi lities and produces realistic full-body m...\n\n\nSebastian Starke and Pau l Starke (Facebook Reality Labs), Taku Komura (University of Hong Kong), a nd Yuting Ye (Facebook Reality Labs)\n---------------------\nTarget-aware Image Denoising for Inverse Monte Carlo Rendering\n\nWe present a novel im age denoiser to improve the convergence of inverse rendering optimization, which infers scene parameters by matching a rendering image to a user-spe cified target image. We reformulate a regression-based denoiser using the target image to make the optimization with our denoising ...\n\n\nJeongmin Gu and Jonghee Back (Gwangju Institute of Science and Technology), Sung-E ui Yoon (Korea Advanced Institute of Science and Technology (KAIST)), and Bochang Moon (Gwangju Institute of Science and Technology)\n-------------- -------\nA Fully-correlated Anisotropic Micrograin BSDF Model\n\nWe introd uce an improved micrograin model for the rendering of porous layers. From the microscopic surface properties, we derived the exact correlation betwe en heights and normals. Our analytical Geometrical Attenuation Factor acco unts for correlation between light and view directions, yielding to r...\n \n\nSimon Lucas (University of Bordeaux; INRIA, Bordeaux Sud-Ouest); Micka ël Ribardière (University of Poitiers - XLIM); and Romain Pacanowski and P ascal Barla (INRIA, Bordeaux Sud-Ouest)\n---------------------\nA Linear M ethod to Consistently Orient Normals of a 3D Point Cloud\n\nConsistently o rienting the normals of a point cloud is vital for subsequent geometry pro cessing. We use Stokes' theorem to turn this problem into finding the leas t-squares solution of a sparse linear system. Our method successfully orie nts normals computed locally from a point cloud and is considera...\n\n\nC raig Gotsman (New Jersey Institute of Technology (NJIT)) and Kai Hormann ( Università della Svizzera Italiana)\n---------------------\nA Vortex Parti cle-on-mesh Method for Soap Film Simulation\n\nWe introduce a sophisticate d vortex particle method for precise tangential flow simulations on membra nes. It uniquely splits membrane velocity into circulation and expansion, employing a hybrid particle-mesh technique to integrate surfactant and thi ckness dynamics. This method accurately captures co...\n\n\nNingxiao Tao ( Yuanpei College, Peking University); Liangwang Ruan (School of CS & Nation al Key Lab of General AI, Peking University); Yitong Deng (Stanford Univer sity); Bo Zhu (Georgia Institute of Technology); Bin Wang (Beijing Institu te for General Artificial Intelligence, State Key Laboratory of General Ar tificial Intelligence); and Baoquan Chen (School of Intelligence Science a nd Technology, Peking University)\n---------------------\nHAISOR: Human-Aw are Indoor Scene Optimization via Deep Reinforcement Learning\n\nHAISOR pr oposes a pipeline to use deep reinforcement learning and Monte Carlo tree search to solve indoor scene optimization problem incorporating human beha vior including human-furniture interaction and free space of activities th at is not differentiable.\n\n\nJia-Mu Sun (Insititute of Computing Technol ogy Chinese Academy of Sciences, University of Chinese Academy of Sciences ); Jie Yang (Chinese Academy of Sciences Institute of Computing Technology ); Kaichun Mo (NVIDIA Research); Yukun Lai (Cardiff University); Leonidas Guibas (Stanford University); Lin Gao (University of the Chinese Academy o f Sciences); and Jia-Mu Sun\n---------------------\nStrategy and Skill Lea rning for Physics-based Table Tennis Animation\n\nWe present a strategy an d skill learning approach for physics-based table tennis animation. We dem onstrate a hierarchical control system for diversified skill learning and a strategy learning framework for effective decision-making. Our strategy learning framework is validated through both agent-age...\n\n\nJiashun Wan g (Carnegie Mellon University); Jungdam Won (Seoul National University); a nd Jessica Hodgins (Carnegie Mellon University, The AI Institute)\n------- --------------\nBiharmonic Coordinates and Their Derivatives for Triangula r 3D Cages\n\nWe provide closed-form expressions for biharmonic coordinate s for triangular cages in 3D, as well as for their gradients and Hessians. We demonstrate new cage-based deformations, a novel subspace for variatio nal cage-based modeling, and we establish closed-form expressions for Somi gliana coordinates...\n\n\nJean-Marc THIERY and Elie Michel (Adobe Researc h) and Jiong Chen (INRIA, Saclay)\n---------------------\nNeuralVDB: High- resolution Sparse Volume Representation using Hierarchical Neural Networks \n\nNeuralVDB enhances the VDB framework for efficient sparse volumetric d ata storage by integrating machine learning. This new structure significan tly reduces memory usage while maintaining flexibility with minimal compre ssion errors. It combines a shallow VDB tree with hierarchical neural netw orks for...\n\n\nDoyub Kim, Minjae Lee, and Ken Museth (NVIDIA, USA) and D oyub Kim\n---------------------\nTerrain Amplification Using Multi Scale E rosion\n\nWe present a multi-scale amplification method producing a high-r esolution, hydrologically consistent terrain. The method relies on a fast and accurate approximation of different erosion processes, including therm al, stream power erosion and deposition performed at different scales to o btain a range ...\n\n\nHugo Schott (CNRS - LIRIS; INSA, Lyon); Eric Galin (CNRS - LIRIS, Université Claude Bernard Lyon 1); Eric Guérin (CNRS - LIRI S; INSA, Lyon); Axel Paris (Adobe Research); and Adrien Peytavie (CNRS - L IRIS, Université Claude Bernard Lyon 1)\n---------------------\nReal-time Physically Guided Hair Interpolation\n\nLinear interpolation can often cau se severe deformation artifacts in hair for which alternatives often entai l expensive training or precomputation. Instead, we present a novel force- based hair interpolation scheme to leverage existing real-time simulation data for a robust interpolation with very li...\n\n\nJerry Hsu (University of Utah, LightSpeed Studios); Tongtong Wang, Zherong Pan, and Xifeng Gao (LightSpeed Studios); Cem Yuksel (University of Utah, Roblox); and Kui Wu (LightSpeed Studios)\n---------------------\nLifting Directional Fields to Minimal Sections\n\nBy lifting direction fields to currents on a circle b undle, we unlock a new convex relaxation of field optimization that treats singularities as first-class citizens.\n\n\nDavid R. Palmer (Harvard Univ ersity), Albert Chern (University of California San Diego), and Justin M. Solomon (Massachusetts Institute of Technology (MIT))\n------------------- --\nStochastic Computation of Barycentric Coordinates\n\nWe introduce an a pproach for computing barycentric coordinates inside and outside a cage do main using only cage queries such as closest points and ray intersections. We show that this stochastic construction reproduces existing barycentric coordinates, e.g., harmonic and (positive) mean-value coordi...\n\n\nFern ando de Goes (Pixar) and Mathieu Desbrun (INRIA, Ecole Polytechnique)\n--- ------------------\nProxy Asset Generation for Cloth Simulation in Games\n \nThis paper presents an automatic pipeline to convert an ill-conditioned high-resolution visual mesh into a single-layer, low-poly proxy mesh. Our approach simulates proxy mesh based on specific use scenarios and optimize s skinning weights, relying on differential skinning with several well-des igned ...\n\n\nZhongtian Zheng (Peking Unversity, LightSpeed Studios) and Tongtong Wang, Qijia Feng, Zherong Pan, Xifeng Gao, and Kui Wu (LightSpeed Studios)\n---------------------\nNeural-assisted Homogenization of Yarn-l evel Cloth\n\nWe introduce a neural-assisted homogenization method for yar n-level cloth. Our approach incorporates a warm-start strategy to enhance the efficiency of the homogenization process. We then use tailored loss fu nctions in constitutive models, ensuring stability for large time-step sim ulations (up to 1/...\n\n\nXudong Feng (State Key Lab of CAD and CG, Zheji ang University; Style3D Research); Huamin Wang (Style3D Research); Yin Yan g (University of Utah, Style3D Research); and Weiwei Xu (State Key Laborat ory of CAD & CG, Zhejiang University)\n---------------------\nEfficient De bris-flow Simulation for Steep Terrain Erosion\n\nDebris flows — flowing m ixtures of mud and rock triggered by extreme climatic events — have a stro ng erosive effect, especially on steep slopes. We propose a new formulatio n and GPU algorithm for the unified modeling of fluvial and debris flow er osion, demonstrating distinct erosion and ...\n\n\nAryamaan Jain (INRIA, U niversité Côte d'Azur); Bedrich Benes (Purdue University); and Guillaume C ordonnier (INRIA, Université Côte d'Azur)\n---------------------\nProxy Tr acing: Unbiased Reciprocal Estimation for Optimized Sampling in BDPT\n\nWe present a novel path sampling technique using path substitution to greatl y enhance BDPT's performance in handling specular or highly glossy involve d paths. We introduce a novel reciprocal estimator along with an efficienc y-optimized setting. This estimator is more efficient and practically appl ic...\n\n\nFujia Su, Bingxuan Li, Qingyang Yin, Yanchen Zhang, and Sheng L i (Peking University)\n---------------------\nGoing With the Flow\n\nGiven a sequence of poses of a body in a fluid medium, we derive generalized ri gid body dynamics equations. These account for fluid inertia and drag-lift forces based on local approximations. This obviates the need for simulati ng the fluid itself and yet yields subtle fluid mediated motion effects.\n \n\nYousuf Soliman (California Institute of Technology); Marcel Padilla, O liver Gross, Felix Knöppel, and Ulrich Pinkall (Technical University of Be rlin); and Peter Schröder (California Institute of Technology)\n---------- -----------\nBinary Opacity Grids: Capturing Fine Geometric Detail for Mes h-based View Synthesis\n\nWe reconstruct meshes from multi-view images tha t contain fine geometric detail yet are suitable for high-quality view syn thesis on low-powered devices. To this end, we model the scene as a high-r esolution binary opacity grid and shoot multiple rays per pixel to be able to reason about subpixel stru...\n\n\nChristian Reiser (Tübingen AI Cente r, University of Tübingen; Google Research); Stephan Garbin, Pratul Sriniv asan, Dor Verbin, Richard Szeliski, Ben Mildenhall, Jonathan Barron, and P eter Hedman (Google Research); and Andreas Geiger (Tübingen AI Center, Uni versity of Tübingen)\n---------------------\nX-SLAM: Scalable Dense SLAM f or Task-aware Optimization Using CSFD\n\nX-SLAM is a real-time dense diffe rentiable SLAM system. It exploits CSFD for an accurate and robust (high-o rder) numerical differentiation, which avoids the need for the computation al graph and substantially reduces the memory footprint. X-SLAM is compati ble with most SLAM algorithms. Evaluations vi...\n\n\nZhexi Peng (Zhejiang University), Yin Yang (University of Utah), Tianjia Shao (Zhejiang Univer sity), Chenfanfu Jiang (University of California Los Angeles), and Kun Zho u (Zhejiang University)\n---------------------\nArea ReSTIR: Resampling fo r Real-time Defocus and Antialiasing\n\nWe introduce Area ReSTIR, extendin g ReSTIR reservoirs to also integrate each pixel’s 4D ray space, including 2D areas on the film and lens. We design novel subpixel-tracking temporal reuse and shift mappings that maximize resampling quality in high-frequen cy regions, thereby allowing efficient...\n\n\nSong Zhang (University of U tah); Daqi Lin, Markus Kettunen, and Chris Wyman (NVIDIA Research); and Ce m Yuksel (University of Utah, Roblox)\n---------------------\nCricket: A S elf-powered Chirping Pixel\n\nCricket is a novel light sensor that can mea sure light without the use of a power supply or a battery. Since it is sel f-powered and untethered, it can be used in a variety of applications incl uding measuring environmental illumination, the control of indoor lighting , and the generation of low-resolu...\n\n\nShree Nayar, Jeremy Klotz, Nikh il Nanda, and Mikhail Fridberg (Columbia University)\n-------------------- -\nRTG-SLAM: Real-time 3D Reconstruction at Scale Using Gaussian Splatting \n\nRTG-SLAM is a real-time 3D reconstruction system using Gaussian splatt ing. It is also memory efficient, enabling reconstruction of large-scale e nvironments. Comparisons demonstrate RTG-SLAM runs at around twice the spe ed of the state-of-the-art, NeRF-based SLAM, with around half the memory c ost (e.g...\n\n\nZhexi Peng, Tianjia Shao, Yong Liu, and Jingke Zhou (Zhej iang University); Yin Yang (University of Utah); Jingdong Wang (Baidu Rese arch); and Kun Zhou (Zhejiang University)\n---------------------\nThe Olym pus Programming Language\n\nOlympus brings a new approach to esolanging, b orrowing from Oulipo and digital literature. It blurs the line between nat ural language and code. Its vocabulary influences its logic, rather than t he reverse. It shows the potential of esolangs as full expressions of arti stic vision. It is written in am...\n\n\nDaniel Temkin (Studio Daniel Temk in)\n---------------------\nLightning-fast Method of Fundamental Solutions \n\nThis work introduces a variational preconditioner, based on the invers e Cholesky factorization, to improve the efficiency of solving dense syste ms discretized from boundary integral equations, effectively addressing th e scalability issue commonly encountered in boundary-based approaches.\n\n \nJiong Chen (INRIA, Saclay); Florian Schaefer (Georgia Institute of Techn ology); and Mathieu Desbrun (INRIA, Saclay)\n---------------------\nPath-s pace Differentiable Rendering of Implicit Surfaces\n\nWe extend the theory of path-space differentiable rendering to include implicit surfaces and p ropose Monte\nCarlo estimators for the boundary integrals. This widens the application of physics-based differentiable\nrendering and opens up new p ossibilities for inverse rendering applications.\n\n\nSiwei Zhou (Osaka Un iversity, Tokyo City University); Youngha Chang and Nobuhiko Mukai (Tokyo City University); Hiroaki Santo, Fumio Okura, and Yasuyuki Matsushita (Osa ka University); and Shuang Zhao (University of California Irvine, NVIDIA)\ n---------------------\nIntrinsicDiffusion: Joint Intrinsic Layers From La tent Diffusion Models\n\nEstimating intrinsic images, such as albedo, shad ing, and normals, is challenging. We propose leveraging the implicit prior s learned by the large-scale generation model. Our novel conditioning mech anism allows predicting multiple intrinsic modalities jointly and training with mixed datasets that onl...\n\n\nJundan Luo (University of Bath); Duy gu Ceylan, Jae Shin Yoon, Nanxuan Zhao, Julien Philip, and Anna Frühstück (Adobe Research); Wenbin Li (University of Bath); Christian Richardt (Meta ); and Tuanfeng Wang (Adobe Research)\n---------------------\nContact Dete ction Between Curved Fibres: High Order Makes a Difference\n\nWhen simulat ing fibre assemblies with standard techniques, we identify spurious artifa cts in the contact forces which are caused by the use of low-order proxys for collision detection. We fix this issue by developing an efficient and accurate high-order detection scheme between two smooth curves, wh...\n\n\ nOctave Crespel, Emile Hohnadel, Thibaut Metivet, and Florence Bertails-De scoubes (University Grenoble Alpes Inria, CNRS, Grenoble INP, LJK)\n------ ---------------\nA Dynamic Duo of Finite Elements and Material Points\n\nD ynamicDuo is a novel framework designed to integrate FEM and MPM seamlessl y. The IMEX framework combines the optimal performance of implicit FEM and the flexibility of explicit MPM in applying plasticity. We achieve this t hrough asynchronous time-splitting, where IPC is applied to model inter-do mai...\n\n\nXuan Li (University of California Los Angeles); Minchen Li (Ca rnegie Mellon University); Xuchen Han (Toyota Research Institute); Huamin Wang (Style3D Research); Yin Yang (University of Utah, Style3D Research); and Chenfanfu Jiang (University of California Los Angeles, Style3D Researc h)\n---------------------\n3D Gaussian Blendshapes for Head Avatar Animati on\n\nWe introduce the 3D Gaussian blendshape representation for modeling photorealistic head avatars. The avatar model of an arbitrary expression c an be effectively generated through linear blending of Gaussian blendshape s with the expression coefficients. Compared to state-of-the-art methods, our method...\n\n\nShengjie Ma, Yanlin Weng, Tianjia Shao, and Kun Zhou (Z hejiang University)\n---------------------\nMaPa: Text-driven Photorealist ic Material Painting for 3D Shapes\n\nThis paper aims to generate material s for 3D meshes from text descriptions. We propose to generate segment-wis e procedural material graphs as the appearance representation, which suppo rts high-quality rendering and provides substantial flexibility in editing . Extensive experiments demonstrate superi...\n\n\nShangzhan Zhang (State Key Lab of CAD and CG, Zhejiang University; Ant Group); Sida Peng, Tao Xu, Yuanbo Yang, and Tianrun Chen (Zhejiang University); Nan Xue and Yujun Sh en (Ant Group); Hujun Bao (State Key Lab of CAD and CG, Zhejiang Universit y); Ruizhen Hu (Shenzhen University); and Xiaowei Zhou (State Key Lab of C AD and CG, Zhejiang University)\n---------------------\nSemantic Gesticula tor: Semantics-aware Co-speech Gesture Synthesis\n\nWe introduce Semantic Gesticulator, a novel framework designed to synthesize realistic co-speech gestures with strong semantic correspondence. Semantic Gesticulator fine- tunes an LLM to retrieve suitable semantic gesture candidates from a motio n library. Combined with a novel, GPT-style generative m...\n\n\nZeyi Zhan g (Peking University, School of EECS); Tenglong Ao (School of Computer Sci ence, Peking University); Yuyao Zhang (Renmin University of China); Qingzh e Gao (Shandong University, Peking University); Chuan Lin (Peking Universi ty); and Baoquan Chen and Libin Liu (Peking University, State Key Laborato ry of General Artificial Intelligence)\n---------------------\nDiffSound: Differentiable Modal Sound Rendering and Inverse Rendering for Diverse Inf erence Tasks\n\nWe propose DiffSound, a differentiable sound rendering fra mework for physics-based modal sound synthesis, which consists of an impli cit shape representation, a high-order finite element analysis module, and a differentiable audio synthesizer. Our framework can solve diverse inver se problems, includi...\n\n\nXutong Jin and Chenxi Xu (Peking University); Ruohan Gao (University of Maryland College Park, Stanford University); Ji ajun Wu (Stanford University); and Guoping Wang and Sheng Li (Peking Unive rsity)\n---------------------\nSimplicits: Mesh-free, Geometry-agnostic El astic Simulation\n\nSimplicits is a versatile framework for reduced elasti c simulations of 3D objects in any geometric representation such as gaussi an splats, SDFs, point-clouds, an even medical scans. Our mesh-free, grid- free method utilizes implicit neural fields to construct a physics-aware s ubspace of the object vi...\n\n\nVismay Modi (University of Toronto, NVIDI A); Nicholas Sharp and Or Perel (NVIDIA Research); Shinjiro Sueda (Texas A &M University); and David Levin (University of Toronto, NVIDIA Research)\n ---------------------\nLatent L-systems: Transformer-based Tree Generator\ n\nThe study introduces a Transformer-based deep learning framework for ge nerating 3D tree models using L-systems. Trained on 155k geometries, this neural model replaces manual rule creation, accurately learning and genera ting L-strings. It shows high fidelity in branching patterns and geometric featur...\n\n\nJae Joong Lee, Bosheng Li, and Bedrich Benes (Purdue Unive rsity) and Jae Joong Lee\n---------------------\nLagrangian Covector Fluid With Free Surface\n\nWe present a novel Lagrangian solver for incompressi ble flows, leveraging flow-maps to simulate vortical evolution with partic les. Our approach focuses on using particle trajectories as flow maps, tai loring path integrals along trajectories for complex boundary conditions, and computing physical qua...\n\n\nZhiqi Li, Barnabás Börcsök, Duowen Chen , Yutong Sun, Bo Zhu, and Greg Turk (Georgia Institute of Technology)\n--- ------------------\nQT-Font: High-efficiency Font Synthesis via Quadtree-b ased Diffusion Models\n\nWe propose a novel sparse glyph representation vi a quadtree and an efficient font synthesis method via dual quadtree and di screte diffusion model, QT-Font. QT-Font, compared to existing approaches, can generate high-resolution glyph images with superior quality and more visually pleasing details, me...\n\n\nYitian Liu (Peking University, Wangx uan Institute of Computer Technology) and Zhouhui Lian (Peking University; Wangxuan Institute of Computer Technology, Peking University)\n---------- -----------\nFabricable 3D Wire Art\n\nWe computationally create fabricabl e 3D wire sculptures from various input modalities, including 3D models, i mages, and texts. Our curve design algorithm automatically abstracts the v arious inputs using a pre-trained, vision-language model and ensures fabri cability using geometrical constraints. 3D-...\n\n\nKenji Tojo (University of Tokyo); Ariel Shamir (Reichman University); Bernd Bickel (Institute of Science and Technology Austria (ISTA), ETH Zürich); and Nobuyuki Umetani (University of Tokyo)\n---------------------\n'Sisyphus': A Robotic Vignet te of Systemic Conflict and Democratic Endeavors\n\nThe robotic art instal lation explores the symbiosis of art and technology to probe crucial socie tal issues. It embodies the eternal struggle between suppressive systems a nd the spirit of collective defiance, using a cybernetic narrative to mirr or the complexities of sociopolitical resistance. By int...\n\n\nKa Chi Ch an (Hong Kong Baptist University; Interactive Architecture Lab, University College London (UCL))\n---------------------\nNeLT: Object-oriented Neura l Light Transfer\n\nOur work presents object-oriented neural light transfe r (NeLT), a novel modular neural representation of the dynamic light trans portation between an object and the environment. It enables interactive re ndering with global illumination for dynamic scenes and achieves comparabl e quality to the recent ...\n\n\nChuankun Zheng, Yuchi Huo, Shaohua Mo, Zh ihua Zhong, and Zhizhen Wu (Zhejiang University State Key Lab of CAD&CG); Wei Hua (Zhejiang Lab); Rui Wang and Hujun Bao (Zhejiang University Stat e Key Lab of CAD&CG); and Chuankun Zheng\n---------------------\nDragon's Path: Synthesizing User-centered Flying Creature Animation Paths for Outdo or Augmented Reality Experiences\n\nWe present a novel approach to automat ically generate user-centered flying creature animation paths for outdoor augmented reality experiences. Given a sequence of storyline actions, our approach finds suitable locations for the character to perform its actions , generating a corresponding animation p...\n\n\nMinyoung Kim (George Maso n University), Rawan Alghofaili (University of Texas at Dallas), and Chang yang Li and Lap-Fai Yu (George Mason University)\n---------------------\nS plit-Aperture 2-in-1 Computational Cameras\n\nWe introduce a split-apertur e 2-in-1 computational camera that combines application-specific optical m odulation with conventional imaging into one system. This approach simplif ies complex inverse problems faced by computational cameras, enhances reco nstruction quality, and offers a real-time viewfin...\n\n\nZheng Shi, Ilya Chugunov, Mario Bijelic, Geoffroi Côté, and Jiwoon Yeom (Princeton Univer sity); Qiang Fu, Hadi Amata, and Wolfgang Heidrich (King Abdullah Universi ty of Science and Technology (KAUST)); and Felix Heide (Princeton Universi ty)\n---------------------\nFluid Control With Laplacian Eigenfunctions\n\ nWe introduce a novel physics-based fluid control pipeline using Laplacian Eigenfluids. Utilizing the adjoint method, the derivative computation of the control problem is efficient and easy to formulate. Our method is fast enough to support real-time fluid editing and control, which also natural ly su...\n\n\nYixin Chen (University of Toronto); David Levin (University of Toronto, NVIDIA); and Timothy Langlois (Adobe Research)\n-------------- -------\nCapacitive Touch Sensing on General 3D Surfaces\n\nThe paper prop oses a new method to adapt the common grid of electrodes of a mutual-capac itive sensor to generic 3D surfaces, minimizing the number of controllers and input/output pins. The tested prototypes show precise and robust multi -touch detection with excellent Signal-to-Noise Ratio and spatia...\n\n\nG ianpaolo Palma (CNR ISTI), Narges Pourjafarian and Jürgen Steimle (Saarlan d Informatics Campus), and Paolo Cignoni (CNR ISTI)\n--------------------- \nMotion-I2V: Consistent and Controllable Image-to-video Generation With E xplicit Motion Modeling\n\nPlease see our anonymous project page for more results: \url{https://anonymous-2718281.github.io}.\n\n\nXiaoyu Shi, Zhaoy ang Huang, Fu-Yun Wang, Weikang Bian, and Dasong Li (The Chinese Universit y of Hong Kong); Yi Zhang (SenseTime); Manyuan Zhang (The Chinese Universi ty of Hong Kong); Ka Chun Cheung and Simon See (NVIDIA); Hongwei Qin (Sens eTime); Jifeng Dai (Tsinghua University); and Hongsheng Li (The Chinese Un iversity of Hong Kong)\n---------------------\nSplit-and-Fit: Learning B-R eps via Structure-aware Voronoi Partitioning\n\nThe "Split-and-Fit" method introduces a top-down, structure-aware strategy for reconstructing B-Rep models, using Voronoi diagrams to partition the space and followed by prim itive fitting. We design NVD-Net to accurately predict these partitions fr om point clouds or distance fields, resulting in sig...\n\n\nYilin Liu (Sh enzhen University, Simon Fraser University); Jiale Chen and Shanshan Pan ( Shenzhen University); Daniel Cohen-Or (Tel Aviv University, Shenzhen Unive rsity); Hao Zhang (Simon Fraser University); and Hui Huang (Shenzhen Unive rsity)\n---------------------\nVertex Block Descent\n\nWe introduce an eff icient and unconditionally stable solution for the variational form of imp licit time integrator through vertex-level position updates with Gauss-Sei del iterations. It can achieve numerical convergence and provides superior performance to alternative techniques for elastic body dyn...\n\n\nAnka H . Chen, Ziheng Liu, and Yin Yang (University of Utah) and Cem Yuksel (Univ ersity of Utah, Roblox)\n---------------------\nConceptLab: Creative Conce pt Generation using VLM-Guided Diffusion Prior Constraints\n\nThe surge of personalization techniques has allowed us to imagine how existing concept s would look in new scenes. However, an intriguing question remains: How c an we generate a new, imaginary concept that has never been seen before? W e propose an approach for creative concept generation using Diffus...\n\n\ nElad Richardson, Kfir Goldberg, Yuval Alaluf, and Daniel Cohen-Or (Tel Av iv University) and Yuval Alaluf\n---------------------\nQuad-optimized Low -discrepancy Sequences\n\nThis paper introduces multi-dimensional low-disc repancy sequences, in base 3 instead of the usual base 2, optimized over g roups of 4 consecutive dimensions, while keeping good uniformity for non-o ptimized n-tuples of dimensions. We improve over Sobol' sequences in terms of sample uniformity in 2D an...\n\n\nVictor Ostromoukhov, David Coeurjol ly, Nicolas Bonneel, and Jean-Claude Iehl (University of Lyon 1, CNRS - LI RIS)\n---------------------\nKinetic Simulation of Turbulent Multifluid Fl ows\n\nWe propose a LBM-based simulation of separated multiphase flows. Ou r use of HOME-LBM encoded velocity-based distributions offers a fast, accu rate, and low-memory solver enabling efficient turbulent multiphase simula tions of miscible, immiscible, or even partially miscible fluids.\n\n\nWei Li and Kui Wu (LightSpeed Studios) and Mathieu Desbrun (INRIA, Ecole Poly technique)\n---------------------\nVR-GS: A Physical Dynamics-aware Intera ctive Gaussian Splatting System in Virtual Reality\n\nWe introduce VR-GS, a physics-aware, interactive VR system for immersive manipulation of 3D co ntent represented with Gaussian Splatting. We utilize XPBD simulation to e nsure real-time interactivity, where Gaussian kernels are deformed by cage meshes using two-level embeddings. Our system integrates ...\n\n\nYing Ji ang (University of Hong Kong, University of California Los Angeles); Chang Yu, Tianyi Xie, and Xuan Li (University of California Los Angeles); Yutao Feng (University of Utah, Zhejiang University); Huamin Wang (Style3D Rese arch); Minchen Li (Carnegie Mellon University); Henry Lau (University of H ong Kong); Feng Gao (Amazon); Yin Yang (University of Utah, Style3D Resear ch); and Chenfanfu Jiang (University of California Los Angeles, Style3D Re search)\n---------------------\nCLAY: A Controllable Large-scale Generativ e Model for Creating High-quality 3D Assets\n\nDiscover CLAY, the innovati ve 3D generative tool designed to transform imagination into digital form with ease. Powered by a 1.5 billion-parameter model, CLAY excels in creati ng high-quality, realistic 3D assets, enabling experts and novices alike t o reignite their creative spark and bring vibrant i...\n\n\nLongwen Zhang, Ziyu Wang, Qixuan Zhang, and Qiwei Qiu (ShanghaiTech University, Deemos T echnology); Anqi Pang and Haoran Jiang (ShanghaiTech University); Wei Yang (Huazhong University of Science and Technology); and Lan Xu and Jingyi Yu (ShanghaiTech University)\n---------------------\nGEM3D: GEnerative Media l Abstractions for 3D Shape Synthesis\n\nWe introduce GEM3D — a deep, topo logy-aware model for generating and reconstructing 3D shapes. Our key ingr edient is a neural skeleton-based representation compactly encoding both s hape topology and geometry. Experiments show significantly more faithful s urface reconstruction and diverse shape...\n\n\nDmitrii Petrov, Pradyumn G oyal, and Vikas Thamizharasan (University of Massachusetts Amherst); Vladi mir Kim and Matheus Gadelha (Adobe Research); Melinos Averkiou (CYENS - Ce ntre of Excellence, University of Cyprus); Siddhartha Chaudhuri (Adobe Res earch); and Evangelos Kalogerakis (University of Massachusetts Amherst, CY ENS - Centre of Excellence)\n---------------------\n4D-Rotor Gaussian Spla tting: Towards Efficient Novel View Synthesis for Dynamic Scenes\n\nWe pre sent 4D-Rotor Gaussian Splatting that represents dynamic scenes with aniso tropic 4D XYZT Gaussians and rotor-based 4D rotations for novel-view synth esis of dynamic scenes. The proposed method outperforms prior arts in rend ering quality and achieves 277 FPS inference speed on an RTX 3090 GPU at.. .\n\n\nYuanxing Duan (Peking University); Fangyin Wei (Princeton Universit y); Qiyu Dai (Peking University, State Key Laboratory of General Artificia l Intelligence); Yuhang He (Peking University); Wenzheng Chen (NVIDIA, Pek ing University); and Baoquan Chen (Peking University, State Key Laboratory of General Artificial Intelligence)\n---------------------\nMonoGaussianA vatar: Monocular Gaussian Point-based Head Avatar\n\nMonoGaussianAvatar ha rnesses 3D Gaussian representation coupled with a Gaussian deformation fie ld to learn explicit head avatars from monocular portrait videos. These Ga ussian points with adaptable shapes can exhibit movement with a Gaussian d eformation field in alignment with the target pose and ex...\n\n\nYufan Ch en (Harbin Institute of Technology), Lizhen Wang and Qijing Li (Tsinghua U niversity), Hongjiang Xiao (Communication University of China), Shengping Zhang and Hongxun Yao (Harbin Institute of Technology), and Yebin Liu (Tsi nghua University)\n---------------------\nSouthern Wind Algorist: Merging Nine Algorithms with Nine Gay Stories Written in Classical Chinese\n\n“Sou thern Wind Algorist,” addressing historical non-Western male homosexuality , brings the content of Chinese Studies and Gay & Lesbian Studies to SIGGR APH, where such artwork has audiences but is rarely presented, creating an opportunity to reflect on an important aspect of human nature...\n\n\nTen gchao Zhou (University of California Los Angeles)\n---------------------\n Training-free Consistent Text-to-image Generation\n\nText-to-image models allow users to generate individual images guided by language. In this work , we enable Stable Diffusion XL (SDXL) to generate consistent subjects acr oss a series of images, without additional training.\n\n\nYoad Tewel (NVID IA Research, Tel Aviv University); Omri Kaduri (Independent); Rinon Gal an d Yoni Kasten (NVIDIA Research); Lior Wolf (Tel Aviv University); and Gal Chechik and Yuval Atzmon (NVIDIA Research)\n---------------------\nDiffCAD : Weakly-supervised Probabilistic CAD Model Retrieval and Alignment From a n RGB Image\n\nDiffCAD introduces probabilistic CAD retrieval and alignmen t to an RGB image. It models the disentangled distributions of scene scale , object pose, and shape leveraging diffusion. This enables cross-domain m ulti hypothesis CAD reconstruction, capturing the inherent ambiguities in depth/scale and obj...\n\n\nDaoyi Gao, David Rozenberszki, Stefan Leuteneg ger, and Angela Dai (Technical University of Munich)\n-------------------- -\nHand-Object Interaction Controller (HOIC): Deep Reinforcement Learning for Reconstructing Interactions With Physics\n\nThis paper introduces a ph ysics-based hand-object interaction reconstruction system by leveraging im itation learning. A novel object compensation control technique is propose d to upgrade the simple point contact model to a more physical-plausible s urface contact model, which improves the training st...\n\n\nHaoyu Hu and Xin-yu Yi (School of Software, Tsinghua University); Zhe Cao (Google); and Jun-Hai Yong and Feng Xu (School of Software, Tsinghua University)\n----- ----------------\nTexPainter: Generative Mesh Texturing With Multi-view Co nsistency\n\nWe propose a novel texture generation method with multi-view consistency using a pre-trained 2D diffusion model. Drawing on the princip le of DDIM scheme and its adept prediction of noisy latent. Our method foc uses on explicitly controlling the consistency of texture while preserving the denoise proc...\n\n\nHongkun Zhang (Southeast University); Zherong Pa n (LightSpeed Studios); Congyi Zhang (University of British Columbia, Tran sGP); Lifeng Zhu (Southeast University); and Xifeng Gao (LightSpeed Studio s)\n---------------------\nNeural Control Variates With Automatic Integrat ion\n\nWe present a method that uses arbitrary neural network architecture s as control variates with automatic differentiation to improve Monte Carl o methods. Our approach creates unbiased, low-variance, and numerically st able Monte Carlo estimators for various problem setups. We demonstrate our method's a...\n\n\nZilu Li (Cornell University), Guandao Yang and Qingqin g Zhao (Stanford University), Xi Deng (Cornell University), Leonidas Guiba s (Stanford University), Bharath Hariharan (Cornell University), and Gordo n Wetzstein (Stanford University)\n---------------------\nDreamFont3D: Per sonalized Text-to-3D Artistic Font Generation\n\nThis paper presents a nov el text-to-3D font generation model. It allows text prompt to describe 3D font styles, and script-generated font masks or hand-drawn font layouts to constrain the 3D font structure at multi-views, achieving stunning 3D rep resentation of artistic font and the control of local...\n\n\nXiang Li (Sh andong University); Lei Meng (Shandong University, Shandong Research Insti tute of Industrial Technology); and Lei Wu, Manyi Li, and Xiangxu Meng (Sh andong University)\n---------------------\nAlignment Conditions for NURBS- based Design of Mixed Tension-compression Grid Shells\n\nWe introduce a no vel PDE that can align conjugate stress and curvature nets for shell form- finding. Users can adjust the grid using a reference grid. This allows use rs to design a metal-glass grid shell that can withstand gravity with bend ing-free axial forces and be covered by planar quadrilateral ...\n\n\nMasa aki Miki (University of Tokyo) and Toby Mitchell (Thornton, Tomasetti)\n-- -------------------\nHigh-quality Surface Reconstruction Using Gaussian Su rfels\n\nWe introduce Gaussian surfels, a novel point-based representation that flattens 3D Gaussian ellipsoids into 2D ellipses. This representatio n combines the advantages of the flexible optimization procedure in 3D Gau ssian points and the surface alignment property of surfels, resulting in s uperior perfo...\n\n\nPinxuan Dai (State Key Laboratory of CAD & CG, Zheji ang University); Jiamin Xu (Hangzhou Dianzi University); Wenxiang Xie and Xinguo Liu (State Key Laboratory of CAD & CG, Zhejiang University); Huamin Wang (Style3D Research); and Weiwei Xu (State Key Laboratory of CAD & CG, Zhejiang University)\n---------------------\nSeamless Parametrization in Penner Coordinates\n\nWe introduce a simple and efficient algorithm for se amless parametrization with prescribed angles at singularities and rotatio ns along homology loops. Our algorithm performs exceptionally well on a la rge dataset based on Thingi10k [Zhou and Jacobson 2016] and on a challengi ng smaller dataset of [Myl...\n\n\nRyan Capouellez and Denis Zorin (New Yo rk University)\n---------------------\nTemporally Stable Metropolis Light Transport Denoising Using Recurrent Transformer Blocks\n\nWe propose a lea rning-based denoising method for Metropolis Light Transport (MLT) based on recurrent Transformer blocks. We show that our Transformer architecture c an more effectively resolve the correlation artifacts compared to the blen ding-based approaches used in previous work.\n\n\nChuhao Chen (University of California San Diego), Yuze He (Tsinghua University), and Tzu-mao Li (U niversity of California San Diego)\n---------------------\nRobust Containm ent Queries Over Collections of Rational Parametric Curves via Generalized Winding Numbers\n\nWe extend the theory of generalized winding numbers to unstructured collections of rational parametric curves with a numerically stable algorithm, thereby allowing for robust and accurate containment cl assifications at arbitrary locations for non-watertight and self-intersect ing shapes.\n\n\nJacob Spainhour (University of Colorado Boulder), David G underman (Indiana University School of Medicine), and Kenneth Weiss (Lawre nce Livermore National Laboratory)\n---------------------\nMedia2Face: Co- speech Facial Animation Generation With Multi-modality Guidance\n\nMedia2F ace can generate highly realistic and expressive 3D facial animations from diverse multimedia inputs — audio, text, and images — trained on the larg est ever co-speech 3D facial animation dataset. With Media2Face, avatars c an now embody complex inner emotions with unprecedented fid...\n\n\nQingch eng Zhao, Pengyu Long, and Qixuan Zhang (ShanghaiTech University, Deemos T echnology); Dafei Qin (University of Hong Kong, Deemos Technology); Han Li ang (ShanghaiTech University); Longwen Zhang (ShanghaiTech University, Dee mos Technology); Yingliang Zhang (DGene Digital Technology Co., Ltd.); and Jingyi Yu and Lan Xu (ShanghaiTech University)\n---------------------\nPr omptable Game Models: Text-Guided Game Simulation via Masked Diffusion Mod els\n\nWe propose Promptable Game Models (PGMs), models of games that can be controlled through natural language actions and desired game states. St ates are modeled through a masked diffusion transformer and rendered throu gh a compositional NeRF. Our PGM outperforms existing neural video game si mulators, u...\n\n\nWilli Menapace and Aliaksandr Siarohin (Snap); Stéphan e Lathuilière (Paris Telecom LTCI, Institut Polytechnique de Paris); Panos Achlioptas (Snap); Vladislav Golyanik (Max Planck Institute for Informati cs); Sergey Tulyakov (Snap); Elisa Ricci (Universita degli Studi di Trento Fondazione Bruno Kessler); and Willi Menapace\n---------------------\nVel ocity-based Monte Carlo Fluids\n\nWe present a velocity-based Monte Carlo fluid solver with operator splitting and walk-on-boundary boundary handlin g, which overcomes the limitations of its existing vorticity-based counter part. Our method can readily incorporate various techniques drawn from con ventional non-Monte Carlo methods, suc...\n\n\nRyusuke Sugimoto, Christoph er Batty, and Toshiya Hachisuka (University of Waterloo)\n---------------- -----\nRip-NeRF: Anti-aliasing Radiance Fields With Ripmap-encoded Platoni c Solids\n\nRip-NeRF is an anti-aliasing 3D scene representation that inte grates anisotropic pre-filtering with Platonic solid faces. It captures hi gh-frequency anisotropic details with swift training times, achieving 37.2 3 PSNR on the Blender dataset within 2.6 hours of training. Due to its sim plicity, other t...\n\n\nJunchen Liu (Beihang University); Wenbo Hu (Tence nt AI Lab); Zhuo Yang and Jianteng Chen (Beijing Institute of Technology); Guoliang Wang and Xiaoxue Chen (Tsinghua University); Yantong Cai (Dermat ology Hospital, Southern Medical University); and Huan-ang Gao and Hao Zha o (Tsinghua University)\n---------------------\nTowards Motion Metamers fo r Foveated Rendering\n\nWe demonstrate that foveated rendering may damage motion perception in AR/VR, leading to an under-estimation of cues such as velocity. To mitigate this, we propose the concept of motion metamers; vi deos that are structurally different but induce the same spatial and motio n perception for the human v...\n\n\nTaimoor Tariq and Piotr Didyk (Univer sità della Svizzera Italiana)\n---------------------\nCritical Climate Mac hine: A Visual and Musical Exploration of Climate Misinformation through M achine Learning\n\nPioneering new ways to engage with art, technology, and society, Critical Climate Machine directly addresses current misinformati on issues. It seeks to mitigate its effects by mediating accurate scientif ic content, combining machine learning strategies. In this paper, we aim t o critically examine th...\n\n\nGaëtan Robillard (Université Paris 8, Univ ersité Gustave Eiffel) and Jérôme Nika (Ircam Centre Pompidou)\n---------- -----------\nPractical Error Estimation for Denoised Monte Carlo Image Syn thesis\n\nWe present a practical error estimation technique for denoised M onte Carlo ray tracing, using aggregated estimates of bias and variance to determine the pixel’s squared error distribution. This leads to a novel s topping criterion for denoised Monte Carlo image synthesis, that efficient ly termi...\n\n\nArthur Firmino (Luxion, Technical University of Denmark); Ravi Ramamoorthi (University of California San Diego); Jeppe Revall Frisv ad (Technical University of Denmark); and Henrik Wann Jensen (Luxion)\n--- ------------------\nA Construct-optimize Approach to Sparse View Synthesis Without Camera Pose\n\nWe leverage the 3D Gaussian splatting method to de velop a novel construct-and-optimize method for sparse view synthesis with out camera poses. We demonstrate results on the Tanks and Temples and Stat ic Hikes datasets with as few as three widely spaced views, showing signif icantly better quality than ...\n\n\nKaiwen Jiang, Yang Fu, Mukund Varma T , Yash Belhe, Xiaolong Wang, Hao Su, and Ravi Ramamoorthi (University of C alifornia San Diego)\n---------------------\nIntrinsic Image Decomposition via Ordinal Shading\n\nWe achieve high-resolution intrinsic decomposition in the wild. Our approach consists of two steps: estimating dense ordinal shading cues, and combining low- and high-resolution ordinal estimations to achieve coherent and detailed shading. Our method allows us to generate dense supervision from mult...\n\n\nChristian Careaga and Yagiz Aksoy (Si mon Fraser University) and Christian Careaga\n---------------------\nAudio Matters Too! Enhancing Markerless Motion Capture With Audio Signals for S tring Performance Capture\n\nMotion Capture of musical instrument performa nce is challenging even with markers. By extracting playing cues inherent in the audio for markerless video motion capture, our method recovers subt le finger-string contacts and intricate playing movements. We further cont ribute the first large-scale Stri...\n\n\nYitong Jin, Zhiping Qiu, and Yi Shi (Central Conservatory of Music, Tsinghua University); Shuangpeng Sun ( Tsinghua University); Chongwu Wang and Donghao Pan (Central Conservatory o f Music); Jiachen Zhao (Tsinghua University); Zhenghao Liang (Weilan Tech) ; Yuan Wang, Xiaobing Li, and Feng Yu (Central Conservatory of Music); and Tao Yu and Qionghai Dai (Tsinghua University)\n---------------------\nPer ceptual Evaluation of Steered Retinal Projection\n\nThis study introduces the first perceptual testbed for Steered Retinal Projection (SRP), a displ ay technology that combines retinal projection and pupil steering. We inve stigate the trade space for SRP viewing experience among eye dynamics, eye -tracking, and pupil steering. We also present a detecti...\n\n\nSeungjae Lee, Seung-Woo Nam, Kevin Rio, Renate Landig, Hsien-Hui Cheng, Lu Lu, and Barry Silverstein (Meta Reality Labs)\n---------------------\nSingular Fol iations for Knit Graph Design\n\nWe build a knit planning framework based on stripe patterns viewed as singular foliations. This perspective allows precise control of striping level sets and automatic singularity matching, preventing any helicing of course rows, a challenging problem for knit gr aph creation. Our framework also hand...\n\n\nRahul Mitra and Erick Jimene z Berumen (Boston University), Megan Hofmann (Northeastern University), an d Edward Chien (Boston University)\n---------------------\nTele-Aloha: A T elepresence System With Low-budget and High-authenticity Using Sparse RGB Cameras\n\nTele-Aloha is a low-budget and high-quality bidirectional telep resence system targeting peer-to-peer communication. Integrating efficient stereo matching and 3DGS-based neural rendering, Tele-Aloha utilizes only four RGB cameras, one consumer-grade GPU, and one autostereoscopic screen to achieve hig...\n\n\nHanzhang Tu and Ruizhi Shao (Tsinghua University); Xue Dong (BOE Technology Group); Shunyuan Zheng (Harbin Institute of Tech nology); Hao Zhang, Lili Chen, Meili Wang, Wenyu Li, and Siyan Ma (BOE Tec hnology Group); Shengping Zhang (Harbin Institute of Technology); and Boya o Zhou and Yebin Liu (Tsinghua University)\n---------------------\nHologra phic Parallax Improves 3D Perceptual Realism\n\nHolographic displays offer promising solutions for future VR/AR display platforms. In this study, we investigate the perceptual implications of various CGH algorithms. Our re sults reveal that the inclusion of parallax cues significantly enhances pe rceptual realism and will help guide the community t...\n\n\nDongyeon Kim and Seung-Woo Nam (Seoul National University), Suyeon Choi (Stanford Unive rsity), Jong-Mo Seo (Seoul National University), Gordon Wetzstein (Stanfor d University), and Yoonchan Jeong (Seoul National University)\n----------- ----------\nN-BVH: Neural Ray Queries With Bounding Volume Hierarchies\n\n N-BVH, a compressed neural architecture, enables efficient ray queries in rendering. Our method seamlessly integrates neural ray queries into standa rd pipelines. By optimizing parameters through an adaptive BVH-driven prob ing scheme, N-BVH can serve accurate ray queries from a compact representa tion...\n\n\nPhilippe Weier (Saarland Informatics Campus, DFKI); Alexander Rath (DFKI, Saarland Informatics Campus); Élie Michel and Iliyan Georgiev (Adobe); Philipp Slusallek (DFKI); and Tamy Boubekeur (Adobe)\n---------- -----------\nWoven Fabric Capture With a Reflection-transmission Photo Pai r\n\nWe expend previous woven fabric capture pipeline to support transmiss ion. For that, we propose a two-layer BSDF model including a new azimuthal ly invariant phase function that matches multiple scattering of real fabri cs well. Then we take reflection-transmission photo pair using simple setu p, recove...\n\n\nYingjie Tang and Zixuan Li (Nankai University); Milos Ha san (Adobe Research); Jian Yang (Nanjing University of Science and Technol ogy); and Beibei Wang (Nanjing University, Nankai University)\n----------- ----------\nComputational Homogenization for Inverse Design of Surface-bas ed Inflatables\n\nSurface-based inflatables are composed of two sheet mate rials joined along selected fusing curves, gaining stiffness when inflated to deploy into bending-active shells. We present a computational framewor k employing numerical homogenization and physics-based simulation for opti mizing over arbitrary ...\n\n\nYingying Ren (EPFL); Julian Panetta (Univer sity of California Davis); and Seiichi Suzuki, Uday Kusupati, Florin Isvor anu, and Mark Pauly (EPFL)\n---------------------\nCNS-Edit: 3D Shape Edit ing via Coupled Neural Shape Optimization\n\nThis paper introduces a new a pproach based on a coupled representation and a neural volume optimization to perform 3D shape editing in latent space. Besides, we provide a set of operators, i.e., copy, delete, resize, and drag, for the users to achieve semantic-aware and high-fidelity shape editing.\n\n\nJingyu Hu, Ka-Hei Hu i, and Zhengzhe Liu (The Chinese University of Hong Kong); Hao Zhang (Simo n Fraser University, Amazon); and Chi-Wing Fu (The Chinese University of H ong Kong)\n---------------------\nLGTM: Local-to-Global Text-driven Human Motion Diffusion Model\n\nWe introduce LGTM, a novel Local-to-Global pipel ine for Text-to-Motion generation based on diffusion model. It decomposes motion description to body-part level with LLMs and encodes them with cor responding body part motion individually, then optimizes whole body motion by attention encoder. As a re...\n\n\nHaowen Sun and Ruikun Zheng (Shenzh en University), Haibin Huang (Kuaishou Technology), Chongyang Ma (ByteDanc e Inc.), and Hui Huang and Ruizhen Hu (Shenzhen University)\n------------- --------\nSelf-Supervised High Dynamic Range Imaging: What Can Be Learned from a Single 8-bit Video?\n\nWe present a self-supervised approach that g enerates an HDR video from a single input SDR video without requiring HDR/ SDR datasets for training. Results are comparable and frequently better th an other state-of-the-art methods.\n\n\nFrancesco Banterle (ISTI CNR); Dem etris Marnerides, Thomas Bashford-Rogers, and Kurt Debattista (University of Warwick WMG, Warwick Manufacturing Group); and Francesco Banterle\n--- ------------------\nPhysical Non-inertial Poser (PNP): Modeling Non-inerti al Effects in Sparse-inertial Human Motion Capture\n\nWe propose a novel a pproach for human motion capture using sparse inertial sensors, considerin g the non-inertial behavior of the root frame during acceleration and rota tion. By modeling fictitious forces and using a novel IMU measurement synt hesis technique, accurate motion capture is achieved with ...\n\n\nXinyu Y i (Tsinghua University), Yuxiao Zhou (ETH Zürich), and Feng Xu (Tsinghua U niversity)\n---------------------\nImplicit Swept Volume SDF: Enabling Con tinuous Collision-free Trajectory Generation for Arbitrary Shapes\n\nOur i nnovative approach to trajectory generation seamlessly navigates the compl exities of continuous collision avoidance for objects in challenging envir onments. Leveraging the Swept Volume Signed Distance Field, our hierarchic al pipeline outperforms traditional methods by deftly handling non-convex. ..\n\n\nJingping Wang and Tingrui Zhang (Zhejiang University); Qixuan Zhan g and Chuxiao Zeng (ShanghaiTech University, Deemos Technology); Jingyi Yu (ShanghaiTech University); Chao Xu (Zhejiang University); Lan Xu (Shangha iTech University); and Fei Gao (Zhejiang University)\n-------------------- -\nMob-FGSR: Frame Generation and Super Resolution for Mobile Real-time Re ndering\n\nWe propose Mob-FGSR, a lightweight supersampling framework desi gned specifically for mobile devices, integrating frame generation with su per-resolution to significantly enhance real-time rendering performance. T his method operates efficiently without high-end GPUs or dedicated hardwar e, surpassing o...\n\n\nSipeng Yang, Qingchuan Zhu, and Junhao Zhuge (Stat e Key Laboratory of CAD & CG, Zhejiang University); Qiang Qiu, Chen Li, Yu zhong Yan, and Huihui Xu (OPPO Computing & Graphics Research Institute); L ing-Qi Yan (University of California Santa Barbara); and Xiaogang Jin (Zhe jiang University; State Key Lab of CAD&CG, Zhejiang University)\n--------- ------------\nLight Codes for Fast Two-Way Human-Centric Visual Communicat ion\n\nLight Codes enable fast and fluid exchange of short messages among users with minimal physical and cognitive effort. Light Codes are based on transmitting and receiving temporal codes using compact and low-cost tran sceiver devices. We design coding techniques, hardware prototypes and appl ications th...\n\n\nMohit Gupta (University of Wisconsin-Madison); Jian Wa ng, Karl Bayer, and Shree Nayar (Snap Research); and Jian Wang\n---------- -----------\nReach for the Arcs: Reconstructing Surfaces From SDFs via Tan gent Points\n\nWe introduce an algorithm to reconstruct a mesh from discre te signed distance function (SDF) samples. We use the information containe d in the SDF to construct an oriented point cloud that is then converted i nto a triangle mesh. Our method has no restrictions on topology.\n\n\nSilv ia Sellán (University of Toronto), Yingying Ren (EPFL), Christopher Batty (University of Waterloo), and Oded Stein (University of Southern Californi a)\n---------------------\nDiffusion Illusions: Hiding Images in Plain Sig ht\n\nDiffusion Illusions: Hiding Images In Plain Sight — a novel pipeline for computationally generating special images that produce secret new ima ges when physically arranged in the real world. Training-free, it adapts a frozen text-to-image diffusion model to optimize these images using score -dis...\n\n\nRyan Burgert, Xiang Li, Abe Leite, and Kanchana Ranasinghe (S tony Brook University) and Michael Ryoo (Stony Brook University, Google)\n ---------------------\nCross-image Attention for Zero-shot Appearance Tran sfer\n\nOur zero-shot appearance transfer approach leverages the semantic understanding of text-to-image models to transfer visual appearance across semantically similar objects. Using our cross-image attention, we establi sh semantic correspondences between two images, combining one image's stru cture with a...\n\n\nYuval Alaluf, Daniel Garibi, Or Patashnik, Hadar Aver buch-Elor, and Daniel Cohen-Or (Tel Aviv University)\n-------------------- -\nCompressed Skinning for Facial Blendshapes\n\nWe present a new skinning decomposition method and its application to facial animation, specificall y blendshape compression. Our new algorithm based on Adam with projection outperforms SOTA (Dem Bones) and even allows us to sparsify the bone-bone transformations for further compression.\n\n\nLadislav Kavan, John Doubles tein, Martin Prazak, Matthew Cioffi, and Doug Roble (Meta)\n-------------- -------\nPosition-based Nonlinear Gauss-Seidel for Quasistatic Hyperelasti city\n\nWe show that a position-based, rather than constraint-based, nonli near Gauss-Seidel approach resolves a number of issues with PBD, particula rly in the quasistatic setting. Our approach retains the essential PBD fea ture of stable behavior with constrained computational budgets but also al lows for con...\n\n\nYizhou Chen and Yushan Han (University of California Los Angeles, Epic Games); Jingyu Chen (University of California Los Angele s); Zhan Zhang (University of California Davis); Alex McAdams (Epic Games) ; and Joseph Teran (University of California Davis, Epic Games)\n--------- ------------\nConsistent Point Orientation for Manifold Surfaces via Bound ary Integration\n\nThis paper introduces a robust method for generating gl obally consistent normals for manifold point clouds, leveraging boundary i ntegration and the harmonic properties of the generalized winding number f ield, demonstrating superiority and robustness.\n\n\nWeizhou Liu, Xingce W ang, Haichuan Zhao, Xingfei Xue, and Zhongke Wu (Beijing Normal University ); Xuequan Lu (La Trobe University); and Ying He (Nanyang Technological Un iversity)\n---------------------\nLite2Relight: 3D-aware Single Image Port rait Relighting\n\nLite2Relight presents a novel approach for achieving 3D consistent viewpoint editing and relighting in portraits at interactive s peeds. Lite2Relight expands upon the generative capabilities of EG3D with an efficient lighting manipulation in the latent manifold, leveraging a li ghtstage dataset to mod...\n\n\nPramod Rao (Max Planck Institute for Infor matics; Saarbrücken Research Center for Visual Computing, Interaction and Artificial Intelligence (VIA)); Gereon Fox (Max Planck Institute for Infor matics); Abhimitra Meka (Google AR/VR); Mallikarjun B R and Fangneng Zhan (Max Planck Institute for Informatics); Tim Weyrich (Friedrich-Alexander-U niversität Erlangen-Nürnberg (FAU)); Bernd Bickel (ETH Zürich, Institute o f Science and Technology Austria (ISTA)); Hanspeter Pfister (Harvard Unive rsity); Wojciech Matusik (Massachusetts Institute of Technology (MIT)); Mo hamed Elgharib (Max Planck Institute for Informatics); and Christian Theob alt (Max Planck Institute for Informatics; Saarbrücken Research Center for Visual Computing, Interaction and Artificial Intelligence (VIA))\n------- --------------\nStabler Neo-Hookean Simulation: Absolute Eigenvalue Filter ing for Projected Newton\n\nWe introduce a novel eigenvalue filtering stra tegy for projected Newton's method to stabilize the optimization of Neo-Ho okean energy under high Poisson's ratios (near 0.5) and large initial volu me change. Our method only requires a single line of code change, while ac hieving significant improvement ...\n\n\nHonglin Chen (Columbia University ); Hsueh-Ti Derek Liu (Roblox, University of British Columbia); David I.W. Levin (University of Toronto); Changxi Zheng (Columbia University); and A lec Jacobson (University of Toronto, Adobe Research)\n-------------------- -\nreconFIGURE: Confronting Audiences with Digital Doppelgängers\n\nThis w ork is significant because it translates cutting-edge technology from rese arch into practical artistic applications. It raises critical questions ab out the aesthetics and ethics surrounding the concept of 'capture'. It als o addresses the challenges of machine vision and related issues, includi.. .\n\n\nFlorian Christoph Bruggisser, Chris Elvis Leisi, Pascal Lund-Jensen , Martin Fröhlich, and Christopher Llyoid Salter (Zurich University of the Arts)\n---------------------\nDifferentiable Geodesic Distance for Intrin sic Minimization on Triangle Meshes\n\nWe present a novel approach for the intrinsic minimization of distance-based objectives defined on triangle m eshes. We demonstrate our differentiable geodesic distance framework on ge odesic networks and membranes on surfaces, two-way coupling between hostin g surface and embedded system, differentiab...\n\n\nYue Li, Logan Numerow, Bernhard Thomaszewski, and Stelian Coros (ETH Zürich)\n------------------ ---\nJoint Stroke Tracing and Correspondence for 2D Animation\n\nWe propos e a joint stroke tracing and correspondence approach to facilitate automat ic inbetweening and 2D animation. Given consecutive raster keyframes along with a single vector image of the starting frame as a guidance, the appro ach generates vector drawings for the remaining keyframes while ensur...\n \n\nHaoran Mo, Chengying Gao, and Ruomei Wang (Sun Yat-sen University) and Haoran Mo\n---------------------\nProbabilistic Craft- Materialization of generated images using digital and traditional craft.\n\nThis work presen ts a novel approach to creating physical artwork with image generators by opening a new territory for exploration which combines digital and physica l. It lays a conceptual framework as well as a practical workflow for craf tspeople to utilize the exciting new opportunities offered by ...\n\n\nSha ran R. Elran (Hebrew University of Jerusalem, Shamoon College of Engineeri ng) and Amit R. Zoran (Hebrew University of Jerusalem)\n------------------ ---\nDiffPoseTalk: Speech-driven Stylistic 3D Facial Animation and Head Po se Generation via Diffusion Models\n\nDiffPoseTalk introduces a novel diff usion-based system for generating speech-driven facial animations and head poses, featuring example-based style control through contrastive learning . It overcomes the scarcity of 3D talking face data by utilizing reconstru cted 3DMM parameters from a newly develope...\n\n\nZhiyao Sun, Tian Lv, Sh eng Ye, Matthieu Lin, and Jenny Sheng (Tsinghua University); Yu-Hui Wen (B eijing Jiaotong University); Minjing Yu (Tianjin University); and Yong-Jin Liu (Tsinghua University)\n---------------------\nView-Independent Adjoin t Light Tracing for Lighting Design Optimization\n\nWe introduce a method that enables continuous optimization of the configuration of luminaires in a 3d scene via differentiable adjoint light tracing. Users paint illumina tion targets directly onto the 3d scene, which enables interactive gradien t-based optimization in a camera-free way.\n\n\nLukas Lipp, David Hahn, an d Pierre Ecormier-Nocca (Technical University of Vienna); Florian Rist (Ki ng Abdullah University of Science and Technology (KAUST), Technical Univer sity of Vienna); Michael Wimmer (Technical University of Vienna); and Luka s Lipp\n---------------------\nAutomatic Digital Garment Initialization Fr om Sewing Patterns\n\nLeveraging a combination of AI classification, heuri stics, and numerical optimization, we have developed an innovative hybrid system designed for the automatic initialization of digital garments and r equiring minimal user intervention.\n\n\nChen Liu (State Key Lab of CAD an d CG, Zhejiang University; Style3D Research); Weiwei Xu (State Key Lab of CAD and CG, Zhejiang University); Yin Yang (University of Utah, Style3D Re search); and Huamin Wang (Style3D Research)\n---------------------\nFlexSc ale: Modeling and Characterization of Flexible Scaled Sheets\n\nWe present a computational approach for modeling the mechanical behavior of flexible scaled sheet materials — 3D-printed hard scales embedded in a soft substr ate. We propose a contact-aware homogenization approach that distills nati ve-level simulation data into a novel macro-mechanical model c...\n\n\nJua n Sebastian Montes Maestre, Yinwei Du, Ronan Hinchet, Stelian Coros, and B ernhard Thomaszewski (ETH Zürich)\n---------------------\nThemeStation: Ge nerating Theme-aware 3D Assets From Few Exemplars\n\nThemeStation is an ad vanced tool for crafting theme-consistent 3D models. From a few exemplars to a universe of 3D assets, our two-stage framework and dual distillation process ensure a good blend of unity and diversity. Unleash your creativit y with ThemeStation and step into the realm of effortless...\n\n\nZhenwei Wang (City University of Hong Kong), Tengfei Wang (Shanghai Aritificial In telligence Laboratory), Gerhard Hancke (City University of Hong Kong), Ziw ei Liu (Nanyang Technological University), and Rynson W.H. Lau (City Unive rsity of Hong Kong)\n---------------------\nProgressive Dynamics for Cloth and Shell Animation\n\nWe present Progressive Dynamics, a coarse-to-fine, level-of-detail, physics-based animation method and design pipeline that provides rapid (and so practical) coarse-resolution previews of frictional ly contacting thin shell and cloth dynamics with progressive improvement t o much higher resolution ani...\n\n\nJiayi Eris Zhang (Stanford University , Adobe); Doug James (Stanford University); and Danny M. Kaufman (Adobe Re search)\n---------------------\nPEA-PODs: Perceptual Evaluation of Algorit hms for Power Optimization in XR Displays\n\nDisplays are a major battery drain on untethered XR devices. Energy-efficient rendering reduces power c onsumption but sacrifices visual quality. We investigated popular VR displ ay and rendering architectures through our electronic measurement prototyp e and a large-scale perceptual study. Our milliwa...\n\n\nKenneth Chen (Ne w York University, Meta); Thomas Wan, Nathan Matsuda, Ajit Ninan, and Alex andre Chapiro (Meta); and Qi Sun (New York University)\n------------------ ---\nBlockFusion: Expandable 3D Scene Generation Using Latent Tri-plane Ex trapolation\n\nBlockFusion is a 3D diffusion-based model that generates an d expands 3D scenes using unit blocks. The denoising diffusion process all ows high-quality and diverse scene generation and auto-regressive scene ex pansion. BlockFusion generates large, high-quality 3D scenes for indoor an d outdoor scenarios...\n\n\nZhennan Wu (University of Tokyo); Yang Li (Ten cent); Han Yan (Shanghai Jiao Tong University); Taizhang Shang, Weixuan Su n, and Senbo Wang (Tencent); Ruikai Cui (Australian National University); Weizhe Liu (Tencent); Hiroyuki Sato (National Institute of Informatics, Ja pan; University of Tokyo); and Hongdong Li and Pan Ji (Tencent)\n--------- ------------\nStylized Rendering as a Function of Expectation\n\nWe propos e a generalization of the rendering equation that captures both realistic light transport and a class of stylized visuals in the same formulation. T his allows physical effects like glossy reflections and color bleed to aff ect stylized visuals, and vice versa.\n\n\nRex West (Aoyama Gakuin Univers ity) and Sayan Mukherjee (University of Tokyo, Blueqat Inc.)\n------------ ---------\nX-Portrait: Expressive Portrait Animation With Hierarchical Mot ion Attention\n\nWe propose X-Portrait, an innovative conditional diffusio n model tailored for generating expressive and temporally coherent portrai t animation. Specifically, given a single portrait as appearance reference , we aim to animate it with motion derived from a driving video, capturing both highly dynamic ...\n\n\nYou Xie, Hongyi Xu, Guoxian Song, Chao Wang, Yichun Shi, and Linjie Luo (ByteDance Inc.)\n---------------------\nLight Former: Light-oriented Global Neural Rendering in Dynamic Scene\n\nThe gen eration of global illumination in real-time has been a long-standing chall enge in the graphics community. This work presents a neural rendering appr oach inspired by many-lights methods and Transformer model, dubbed LightFo rmer, that can generate realistic global illumination for fully dynamic... \n\n\nHaocheng Ren (State Key Laboratory of CAD & CG, Zhejiang University) ; Yuchi Huo (State Key Laboratory of CAD & CG, Zhejiang University; Zhejia ng Lab); Yifan Peng (University of Hong Kong); and Hongtao Sheng, Weidong Xue, Hongxiang Huang, Jingzhen Lan, Rui Wang, and Hujun Bao (State Key Lab oratory of CAD & CG, Zhejiang University)\n---------------------\nPrimal-d ual Non-smooth Friction for Rigid Body Animation\n\nOur novel contact solv er, based on an primal-dual interior point algorithm, efficiently and robu stly simulates the non-smooth transition between static and dynamic fricti on. Our algorithm is well suited for large systems of tightly packed objec ts with many contacts, which we demonstrate with complex...\n\n\nYi-Lu Che n, Mickaël Ly, and Chris Wojtan (Institute of Science and Technology Austr ia (ISTA))\n---------------------\nVRMM: A Volumetric Relightable Morphabl e Head Model\n\nWe present the Volumetric Relightable Morphable Model (VRM M), a novel volumetric and parametric facial prior. Our VRMM utilizes a no vel training framework to efficiently disentangle and encode identity, exp ression, and lighting into low-dimensional representations. It facilitates the reconstruction ...\n\n\nHaotian Yang, Mingwu Zheng, and Chongyang Ma (Kuaishou Technology); Yu-Kun Lai (Cardiff University); and Pengfei Wan an d Haibin Huang (Kuaishou Technology)\n---------------------\nReal-Time Neu ral Appearance Models\n\nWe present a real-time rendering system for film- quality materials which bakes complex layered material graphs into compact neural representations. We use learned hierarchical textures and neural d ecoders which use graphics priors to produce neural BSDFs. Accelerated ten sor operations facilitate sea...\n\n\nTizian Zeltner, Fabrice Rousselle, A ndrea Weidlich, Petrik Clarberg, Jan Novák, Benedikt Bitterli, Alex Evans, Tomáš Davidovič, Simon Kallweit, and Aaron Lefohn (NVIDIA) and Tizian Zel tner\n---------------------\nZeroGrads: Learning Local Surrogates for Non- differentiable Graphics\n\nZeroGrads learns a mapping between a graphic pi peline's input parameters and the corresponding loss. The gradient of this mapping can then be used for inverse optimization, enabling the optimizat ion of black-box graphics problems.\n\n\nMichael Fischer and Tobias Ritsch el (University College London (UCL))\n---------------------\nGPU-accelerat ed Rendering of Vector Brush Strokes\n\nThis paper introduces GPU-accelera ted rendering techniques for digital painting and animation to bridge the gap between raster and vector stroke representations, including the suppor t for different brush types (vanilla, stamp, airbrush) and an open-source prototype system that can dynamically update...\n\n\nShen Ciao, Zhongyue G uan, and Qianxi Liu (Hong Kong University of Science and Technology, Guang zhou); Li-Yi Wei (Adobe Research); and Zeyu Wang (Hong Kong University of Science and Technology, Guangzhou; Hong Kong University of Science and Tec hnology)\n---------------------\nFlexible Motion In-betweening With Diffus ion Models\n\nWe introduce CondMDI, a simple and unified diffusion-based a pproach for text and keyframe guided motion in-betweening. CondMDI accommo dates flexible keyframe placements and partial keyframes, generating high- quality and diverse motions coherent with the input constraints. Evaluatio n on the HumanML3D...\n\n\nSetareh Cohan (University of British Columbia); Guy Tevet (Tel Aviv University); Daniele Reda (University of British Colu mbia); Xue Bin Peng (Simon Fraser University, NVIDIA Research); and Michie l van de Panne (University of British Columbia)\n---------------------\nLe arning Images Across Scales Using Adversarial Training\n\nWe propose a nov el paradigm to learn a scale space from an unstructured image collection u sing adversarial training. Enabling zoom-in factors of 256x, our approach can be used to train a multiscale generative model and for reconstructions of scale spaces from unstructured patches.\n\n\nKrzysztof Wolski, Adarsh Djeacoumar, Alireza Javanmardi, Hans-Peter Seidel, and Christian Theobalt (Max Planck Institute for Informatics); Guillaume Cordonnier (INRIA, Unive rsité Côte d'Azur); Karol Myszkowski (Max Planck Institute for Informatics ); George Drettakis (INRIA, Université Côte d'Azur); Xingang Pan (Nanyang Technological University); and Thomas Leimkühler (Max Planck Institute for Informatics)\n---------------------\nBrepGen: A B-rep Generative Diffusio n Model With Structured Latent Geometry\n\nBrepGen is a diffusion-based ge nerative approach that directly outputs a Boundary representation (B-rep) CAD model. It uses a novel structured latent geometry to encode the topolo gy and geometry as a hierarchical tree. A top-down latent diffusion sequen tially denoises the faces, edges, vertices, and...\n\n\nXiang Xu (Simon Fr aser University, Autodesk Research); Joseph Lambourne and Pradeep Jayarama n (Autodesk, Autodesk Research); Zhengqing Wang (Simon Fraser University); Karl Willis (Autodesk, Autodesk Research); and Yasutaka Furukawa (Simon F raser University, Autodesk)\n---------------------\nNICER: A New and Impro ved Consumed Endurance and Recovery Metric to Quantify Muscle Fatigue of M id-air Interactions\n\nNatural gestures are crucial for mid-air interactio ns, but predicting and managing muscle fatigue is challenging. We present a new computational model, NICER, which combines a correction term for 3D interaction and a recovery factor for a rest period to predict fatigue acr oss a variety of gesture-ba...\n\n\nYi Li, Benjamin Tag, and Shaozhang Dai (Monash University); Robert Crowther (University of New England); Tim Dwy er (Monash University); Pourang Irani (The University of British Columbia, Okanagan); and Barrett Ens (The University of British Columbia, Okanagan; Monash University)\n---------------------\nThe Chosen One: Consistent Cha racters in Text-to-image Diffusion Models\n\nGiven a text prompt describin g a character, our method distills a representation that enables consisten t depiction of the same character in novel contexts.\n\n\nOmri Avrahami (H ebrew University of Jerusalem), Amir Hertz (Google), Yael Vinker and Moab Arar (Tel Aviv University), Shlomi Fruchter (Google), Ohad Fried (The Inte rdisciplinary Center Herzliya), Daniel Cohen-Or (Tel Aviv University), and Dani Lischinski (Hebrew University of Jerusalem)\n---------------------\n GaussianPrediction: Dynamic 3D Gaussian Prediction for Motion Extrapolatio n and Free View Synthesis\n\nWe present a novel framework, GaussianPredict ion, for forecasting future scenarios in dynamic scenes. GaussianPredictio n employs a 3D Gaussian canonical space with deformation modeling and life cycle property to represent changes effectively. Additionally, a novel con centric motion distillation techn...\n\n\nBoming Zhao and Yuan Li (State K ey Laboratory of CAD & CG, Zhejiang University); Ziyu Sun (Jilin Universit y); Lin Zeng (State Key Laboratory of CAD & CG, Zhejiang University); Yuju n Shen (Ant Group); Rui Ma (Jilin University); Yinda Zhang (Google); and H ujun Bao and Zhaopeng Cui (State Key Laboratory of CAD & CG, Zhejiang Univ ersity)\n---------------------\nEphemera: Language as a Virus — AI-driven Interactive and Immersive Art Installation\n\nIn this paper, we present th e implementation of "Ephemera," a speech-based interactive and immersive i nstallation that elucidates our artistic exploration of the intersection b etween art, technology, and language. Leveraging multimodal AI technology, we navigate the intricacies of censored language ...\n\n\nJiayang Huang, Yue Huang, David Yip, and Varvara Guljajeva (Hong Kong University of Scien ce and Technology, Guangzhou)\n---------------------\nToonify3D: StyleGAN- based 3D Stylized Face Generator\n\nWe present a StyleGAN-based 3D stylize d face generation framework, Toonify3D, which automatically creates full-h ead 3D stylized avatar mesh models and supports GAN-based 3D facial expres sion editing.\n\n\nWonjong Jang, Yucheol Jung, Hyomin Kim, Gwangjin Ju, Ch aewon Son, Jooeun Son, and Seungyong Lee (POSTECH)\n---------------------\ nDifferentiable solver for time-dependent deformation problems with contac t\n\nWe introduce a general differentiable solver for time-dependent defor mation problems to solve PDE-constrained optimizations. It supports differ entiation with respect to all physical parameters involved, including shap e, material parameters, friction coefficient, and initial conditions. Our analytica...\n\n\nZizhou Huang (New York University); Davi Colli Tozoni (n Top, New York University); Arvi Gjoka (New York University); Zachary Fergu son (Massachusetts Institute of Technology (MIT), New York University); Te seo Schneider (University of Victoria); Daniele Panozzo and Denis Zorin (N ew York University); and Zizhou Huang\n---------------------\nConditional Mixture Path Guiding for Differentiable Rendering\n\nThis work develops a path guiding based on a distribution mixture that improves the performance of differentiable rendering processes. It demonstrates why the mixture is required, how to obtain this distribution mixture, how to update each dis tribution at each iteration, and how to handle negative g...\n\n\nZhimin F an (Nanjing University); Pengcheng Shi (State Key Lab for Novel Software T echnology, Nanjing University); and Mufan Guo, Ruoyu Fu, Yanwen Guo, and J ie Guo (Nanjing University)\n---------------------\nSoft Pneumatic Actuato r Design Using Differentiable Simulation\n\nWe propose a computational des ign pipeline for pneumatically actuated soft robots interacting with their environment through contact. Our approach optimizes the shape of the robo t using a shape optimization approach based on a physically accurate high- order finite element model for forward simulatio...\n\n\nArvi Gjoka (New Y ork University/ Courant), Espen Knoop and Moritz Bächer (Disney Research), and Denis Zorin and Daniele Panozzo (New York University/ Courant)\n----- ----------------\nSpin-It Faster: Quadrics Solve All Topology Optimization Problems That Depend Only on Mass Moments\n\nWe show that a class of topo logy optimization problems used to design statically or rotationally balan ced rigid bodies admits optimal solutions with material interfaces formed by quadrics, such as ellipsoids or hyperboloids. This reduces the number o f unknowns from infinite to less than 10 and yield...\n\n\nChristian Hafne r, Mickaël Ly, and Chris Wojtan (Institute of Science and Technology Austr ia (ISTA))\n---------------------\nSaccade-contingent Rendering\n\nIn this work we introduce saccade-contingent rendering, a perceptually optimized rendering technique which takes advantage of reductions in visual acuity a fter saccades.\n\n\nYuna Kwak (Reality Labs Research, Meta; New York Unive rsity); Eric Penner and Xuan Wang (Reality Labs Research, Meta); Mohammad R. Saeedpour-Parizi (Reality Labs, Meta); Olivier Mercier (Reality Labs Re search, Meta); Xiuyun Wu and Scott Murdison (Reality Labs, Meta); and Phil lip Guan (Reality Labs Research, Meta)\n---------------------\nScale-invar iant Monocular Depth Estimation via SSI Depth\n\nWe propose leveraging SSI depth estimation to achieve highly detailed SI-only depth in the wild. Th is streamlines the network's role and facilitates in-the-wild generalizati on for SI-only depth estimation. Using a synthetic dataset and a novel spa rse ordinal loss, our method generates detailed depth...\n\n\nS. Mahdi H. Miangoleh, Mahesh Reddy, and Yağız Aksoy (Simon Fraser University)\n------ ---------------\nAdaptive Grid Generation for Discretizing Implicit Comple xes\n\nWe present a method for generating a simplicial (e.g., triangular o r tetrahedral) grid to enable adaptive discretization of implicit shapes d efined by a vector function, including the arrangement of implicit surface s, CSG shapes, material interfaces, and curve networks.\n\n\nYiwen Ju and Xingyi Du (Washington University in St. Louis), Qingnan Zhou and Nathan Ca rr (Adobe Research), and Tao Ju (Washington University in St. Louis)\n---- -----------------\nA Framework for Solving Parabolic Partial Differential Equations on Discrete Domains\n\nWe present a method to solve a class of p arabolic PDE on triangle mesh surfaces. Our method leverages a splitting integrator combined with a convex optimization step. We apply our method o n a number of linear and nonlinear PDE that appear in diffusion and front propagation tasks in geometry process...\n\n\nLeticia Mattos Da Silva (Mas sachusetts Institute of Technology (MIT)), Oded Stein (University of South ern California), Justin Solomon (Massachusetts Institute of Technology (MI T)), and Leticia Mattos Da Silva\n---------------------\nDirect-a-Video: C ustomized Video Generation With User-directed Camera Movement and Object M otion\n\nWe introduce Direct-a-Video, a text-to-video generation tool that goes beyond simple text instructions by allowing individual or joint cont rol over camera movements and object motion. It simplifies customized vide o creation, enabling users to craft videos with their preferred motion pat terns, as if ...\n\n\nShiyuan Yang (Tianjin University, City University of Hong Kong); Liang Hou, Haibin Huang, Chongyang Ma, Pengfei Wan, and DI Zh ang (Kuaishou Technology); Xiaodong Chen (Tianjin University); and Jing Li ao (City University of Hong Kong)\n---------------------\nCreating LEGO Fi gurines From Single Images\n\nThis paper introduces a three-stage computat ional pipeline for creating personalized LEGO figurines from portrait phot os: CLIP-guided template-model creation, decal extraction and generation, and physical construction. The resulting figurines not only look like the inputs in the portrait photos with...\n\n\nJiahao Ge, Mingjun Zhou, and We nrui Bao (The Chinese University of Hong Kong); Hao Xu (Qianzhi Technology Inc); and Chi-Wing Fu (The Chinese University of Hong Kong)\n------------ ---------\nSketchDream: Sketch-based Text-to-3D Generation and Editing\n\n We present SketchDream, a novel method that supports text-based NeRF gener ation from given hand-drawn sketches and achieves free-view, sketch-based local editing. We introduce a sketch-based, multi-view image generation di ffusion model to support 3D generation with a coarse-to-fine editing appro ach ...\n\n\nFeng-Lin liu (Institute of Computing Technology, Chinese Acad emy of Sciences; University of Chinese Academy of Sciences); Hongbo Fu (Ci ty University of Hong Kong); Yu-Kun Lai (Cardiff University); and Lin Gao (Institute of Computing Technology, Chinese Academy of Sciences; Universit y of Chinese Academy of Sciences)\n---------------------\nSpecular Polynom ials\n\nA reformulation of specular constraints into polynomial systems th at enables efficiently finding a complete set of all admissible specular p aths connecting two arbitrary endpoints in a scene, by converting the prob lem into finding zeros of the determinant of univariate matrix polynomials .\n\n\nZhimin Fan, Jie Guo, Yiming Wang, and Tianyu Xiao (Nanjing Universi ty); Hao Zhang (Southeast University); Chenxi Zhou and Zhenyu Chen (Nanjin g University); Pengpei Hong (University of Utah); Yanwen Guo (Nanjing Univ ersity); and Ling-Qi Yan (University of California Santa Barbara)\n------- --------------\nAperture-aware Lens Design\n\nWe present a method for desi gning lenses that uses a gradient-based optimization procedure to increase light collection. Our proposed specular warp field enables the differenti ation of light throughput with respect to lens parameters and gives us acc ess to loss functions that explicitly depend on li...\n\n\nArjun Teh, Ioan nis Gkioulekas, and Matthew O'Toole (Carnegie Mellon University)\n-------- -------------\nSuperPADL: Scaling Language-directed Physics-based Control With Progressive Supervised Distillation\n\nWe present a framework for sca ling physics-based text-to-motion models to datasets containing thousands of motions. Our approach being with using reinforcement learning to train a large number of expert tracking policies. We then progressively distill these experts into larger, more capable networks...\n\n\nJordan B. Juravsk y (NVIDIA, Stanford University); Yunrong Guo (NVIDIA); Sanja Fidler (NVIDI A, University of Toronto); and Xue Bin Peng (NVIDIA, Simon Fraser Universi ty)\n---------------------\nTIP-Editor: An Accurate 3D Editor Following Bo th Text-prompts and Image-prompts\n\nWe propose a 3D scene editing framewo rk, TIP-Editor, that accepts both text and image prompts. With the image p rompt, users can conveniently specify the detailed appearance/style of the target content in complement to the text description, enabling accurate c ontrol on the appearance.\n\n\nJingyu Zhuang (Sun Yat-sen University, Tenc ent AI Lab); Di Kang and Yan-Pei Cao (Tencent AI Lab); Guanbin Li and Lian g Lin (Sun Yat-sen University, Peng Cheng Laboratory); and Ying Shan (Tenc ent AI Lab)\n---------------------\n3D Gaussian Splatting With Deferred Re flection\n\nWe present a deferred shading method to effectively render spe cular reflection with Gaussian splatting, achieving high-quality rendering superior to state-of-the-art methods for both synthetic and real-world sc enes. It runs at real-time frame rates almost identical to vanilla Gaussia n splatting and ...\n\n\nKeyang Ye, Qiming Hou, and Kun Zhou (Zhejiang Uni versity)\n---------------------\nA Neural Network Model for Efficient Musc uloskeletal-driven Skin Deformation\n\nWe present a comprehensive neural n etwork pipeline that realistically models soft tissue deformation of anima ted human characters. This network, trained with layered quasi-static simu lation of muscle, fascia, and fat, can effectively replicate musculoskelet al behaviors determined by skeletal kinemat...\n\n\nYushan Han and Yizhou Chen (University of California Los Angeles, Epic Games); Carmichael Ong (S tanford University); Jingyu Chen (University of California Los Angeles); J ennifer Hicks (Stanford University); and Joseph Teran (University of Calif ornia Davis, Epic Games)\n---------------------\nText-guided Synthesis of Crowd Animation\n\nThis paper presents a novel approach to automatically g enerate crowd behaviors from high-level text descriptions. The group-wise, velocity-field-based representation of collective agent behaviors is used and seamlessly integrated with diffusion models and large language models . This work paves the w...\n\n\nXuebo Ji (University of Hong Kong, Centre for Transformative Garment Production (TransGP)); Zherong Pan and Xifeng G ao (Tencent America); and Jia Pan (University of Hong Kong, Centre for Tra nsformative Garment Production (TransGP))\n---------------------\nNavigati on-driven Approximate Convex Decomposition\n\nWe introduce a new approach to approximate convex decomposition based on protecting the space around i nput shapes that objects in a game or simulation must be able to navigate. Our method is efficient, customizable, and provides novel guarantees that the resulting decompositions will meet applicatio...\n\n\nJames Andrews ( Epic Games)\n---------------------\nSpin-weighted Spherical Harmonics for Polarized Light Transport\n\nWe introduce polarized spherical harmonics (P SH), based on spin-weighted spherical harmonics theory, offering a rotatio n-invariant representation of Stokes vector fields. Our approach includes frequency domain formulations of polarized rendering and spherical convolu tion with PSH, making it the firs...\n\n\nShinyoung Yi, Donggun Kim, and J iwoong Na (Korea Advanced Institute of Science and Technology (KAIST)); Xi n Tong (Microsoft Research Asia); and Min H. Kim (Korea Advanced Institute of Science and Technology (KAIST))\n---------------------\nTheory of Huma n Tetrachromatic Color Experience and Printing\n\nWe apply d-dimensional c olor theory to compute the predicted hue sphere of human tetrachromatic co lor space for the first time and derive the ideal, CMY-equivalent, tetrach romatic printing primaries. Our prototype tetrachromatic printer utilizes four selected fountain pen inks to print color tests t...\n\n\nJessica Lee , Nicholas Jennings, Varun Srivastava, and Ren Ng (University of Californi a Berkeley)\n---------------------\nCWF: Consolidating Weak Features in Hi gh-quality Mesh Simplification\n\nIn this paper, we propose an objective f unction that concurrently integrates the requirements of accuracy, triangl e quality, and feature alignment in high-quality mesh simplification. Our function incorporates the normal anisotropy term and the CVT energy term, balanced with a decaying weight.\n\n\nRui Xu and Longdu Liu (Shandong Univ ersity), Ningna Wang (University of Texas at Dallas), Shuangmin Chen (Qing dao University of Science and Technology), Shiqing Xin (Shandong Universit y), Xiaohu Guo (University of Texas at Dallas), Zichun Zhong (Wayne State University), Taku Komura (University of Hong Kong), Wenping Wang (Texas A& M University), and Changhe Tu (Shandong University)\n--------------------- \nST-4DGS: Spatial-Temporally Consistent 4D Gaussian Splatting for Efficie nt Dynamic Scene Rendering\n\nOur ST-4DGS introduces an effective solution to prevent the compactness degeneration of 4D Gaussians by using motion-a ware shape regularization and spatial-temporal joint density control, whic h is the key factor for better dynamic rendering quality in a more efficie nt manner.\n\n\nDeqi Li, Shi-Sheng Huang, Hua Huang, Zhiyuan Lu, and Xinra n Duan (Beijing Normal University)\n---------------------\nScintilla: Simu lating Combustible Vegetation for Wildfires\n\nThis paper introduces a new method for simulating wildfires, integrating detailed vegetation models a nd dynamic interactions like convection and combustion. It realistically d epicts fire progression, ember transport, and the impact of interventions, validated through experiments and real-world data ...\n\n\nAndrzej Kokosz a (AMU), Helge Wrede (Kiel University), Daniel Gonzalez Esparza (King Abdu llah University of Science and Technology (KAUST)), Milosz Makowski (AMU), Daoming Liu and Dominik L. Michels (King Abdullah University of Science a nd Technology (KAUST)), Soren Pirk (Kiel University), and Wojtek Palubicki (AMU)\n---------------------\nImplicit Surface Tension for SPH Fluid Simu lation\n\nIn this paper, we derive and implement an implicit cohesion forc e based approach for the simulation of surface tension effects using the S moothed Particle Hydrodynamics (SPH) method. An adapted variant of the lin earized backward Euler method is used for time discretization, which we st rongly couple ...\n\n\nStefan Rhys Jeske, Lukas Westhofen, Fabian Löschner , José Antonio Fernández-Fernández, and Jan Bender (RWTH Aachen University ) and Stefan Rhys Jeske\n---------------------\nInto the Portal: Directabl e Fractal Self-Similarity\n\nWe present a novel method to create self-simi lar fractals from arbitrary input shapes. Our method introduces "portals" into an iterated map, allowing for user placement of self-similarities and bridging the aesthetics of iterated maps with the fine-grained control of iterated function systems (IFS) ...\n\n\nAlexa Schor and Theodore Kim (Ya le University)\n---------------------\nCybersickness Reduction via Gaze-co ntingent Image Deformation\n\nWe introduce a novel technique to mitigate c ybersickness by gaze-contingent geometrical distortions that reduce vectio n. Based on pre-study insights, we formulate a perceptual model that defin es noticeability limits for these modifications. Our technique, implemente d as a real-time post-processing s...\n\n\nColin Groth, Marcus Magnor, Ste ve Grogorick, and Martin Eisemann (Technical University of Braunschweig) a nd Piotr Didyk (Università della Svizzera Italiana)\n--------------------- \nIterative Motion Editing With Natural Language\n\nWe present a method fo r using natural language to iteratively and conversationally specify local edits to existing character animations. Our key idea is to represent a sp ace of motion edits using a set of operators that have well-defined semant ics for how to modify specific frames of a target motion...\n\n\nPurvi Goe l (Stanford University), Kuan-Chieh Wang (Snap), and C. Karen Liu and Kayv on Fatahalian (Stanford University)\n---------------------\nGenerative Esc her Meshes\n\nThis work achieves a text-guided automatic technique for gen erating visually pleasing 2D tilings similar to the works of M.C. Escher. The method relies on a new parameterization of the space of valid tiles us ing Orbifold Tutte Embeddings and spaces of Laplacians, which enables simp le gradient-based ...\n\n\nNoam Aigerman (University of Montreal) and Thib ault Groueix (Adobe Research)\n---------------------\nLayGA: Layered Gauss ian Avatars for Animatable Clothing Transfer\n\nWe present Layered Gaussia n Avatars (LayGA), a new representation that models layered human avatars based on Gaussians. Our method enables animatable clothing transfer and ac hieves high-fidelity rendering results.\n\n\nSiyou Lin, Zhe Li, and Zhaoqi Su (Tsinghua University); Zerong Zheng (NNKosmos Technology); Hongwen Zha ng (Beijing Normal University); and Yebin Liu (Tsinghua University)\n----- ----------------\nThe Method of Moving Frames for Surface Global Parametri zation\n\nWe introduce a surface parametrization algorithm supporting seam less constraints and feature alignment based on Cartan’s method of moving frames. Using a discretization of Cartan’s structure equations, we derive a non-linear least-square problem which optimizes both singularity positio ns...\n\n\nGuillaume Coiffier and Etienne Corman (Université de Lorraine, Inria, LORIA) and Guillaume Coiffier\n---------------------\nMoConVQ: Unif ied Physics-based Motion Control via Scalable Discrete Representations\n\n We present MoConVQ, a uniform framework enabling simulated avatars to acqu ire diverse skills from large, unstructured datasets. Leveraging a rich an d scalable discrete skill representation, MoConVQ supports a broad range o f applications, including pose estimation, interactive control, text-to-mo tion...\n\n\nHeyuan Yao and Zhenhua Song (School of Computer Science, Peki ng University); Yuyang Zhou (Peking University, School of EECS); Tenglong Ao (School of Computer Science, Peking University); and Baoquan Chen and L ibin Liu (Peking University, State Key Laboratory of General Artificial In telligence)\n---------------------\nCell Space: Augmented Awareness of Int ercorporeality\n\nThis paper presents Cell Space as an artwork that makes awareness of the concept of expanded intercorporeality, moving beyond mere physical interaction to a shared bio-digital domain. It offers a new para digm for applying AR and neurofeedback in contact improvisation, fostering intercorporeal intera...\n\n\nRem RunGu Lin (Hong Kong University of Scie nce and Technology, Guangzhou); Botao Hu (Holo Interactive, Reality Design Lab); Koo Yongen Ke (Fun Theory, BeFun Lab); Wei Wu (BNU-HKBU United Inte rnational College); and Kang Zhang (Hong Kong University of Science and Te chnology, Guangzhou)\n---------------------\n2D Gaussian Splatting for Geo metrically Accurate Radiance Fields\n\n2D Gaussian Splatting (2DGS) improv es upon 3D Gaussian Splatting (3DGS) by converting 3D volumes into 2D Gaus sian disks. It employs perspective-accurate 2D splatting, integrates depth distortion and normal consistency for improved reconstructions, and offer s detailed geometry reconstruction with fas...\n\n\nBinbin Huang (Shanghai Tech University); Zehao Yu, Anpei Chen, and Andreas Geiger (University of Tübingen); and Shenghua Gao (ShanghaiTech University)\n------------------- --\nGIPC: Fast and stable Gauss-Newton optimization of IPC barrier energy\ n\nWe propose a method for rewriting the IPC barrier function using a simp licial-geometric measure of contact. The result is an efficient approximat ion of the Hessian, thanks to our formulation of its analytic eigensystems . These eigensystems further enable an entirely GPU-based IPC system, resu lting i...\n\n\nKemeng Huang, Floyd Chitalu, Huancheng Lin, and Taku Komur a (University of Hong Kong) and Kemeng Huang\n---------------------\nHeadA rtist: Text-conditioned 3D Head Generation With Self Score Distillation\n\ nWe present HeadArtist for 3D head generation and editing following human- language descriptions. With proposed self-score distillation (SSD), we com es up with an efficient pipeline that optimizes a parameterized 3D head mo del under the supervision of a landmark-guided ControlNet itself. Experime ntal ...\n\n\nHongyu Liu (HKUST), Xuan Wang (Ant Group), Ziyu Wan (City Un iversity of Hong Kong), Yujun Shen (Ant Group), Yibing Song (Alibaba DAMO Academy), Jing Liao (City University of Hong Kong), and Qifeng Chen (HKUST )\n---------------------\nDiLightNet: Fine-grained Lighting Control for Di ffusion-based Image Generation\n\nThis paper presents a novel method for e xerting fine-grained lighting control during text-driven, diffusion-based image generation. We demonstrate our lighting-controlled diffusion model o n a variety of text-prompt generated images and under different types of l ighting, ranging from point lights to ...\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); Youkang K ong (Tsinghua University); Hongzhi Wu (State Key Lab of CAD and CG, Zhejia ng University); and Xin Tong (Microsoft Research Asia)\n------------------ ---\nDecorrelating ReSTIR Samplers via MCMC Mutations\n\nWe demonstrate ho w interleaving Markov Chain Monte Carlo mutations with reservoir resamplin g helps alleviate correlation issues in ReSTIR, especially in scenes with glossy materials and difficult lighting. Our approach is free from bias, a nd can provide considerable improvement in image quality with...\n\n\nRoha n Sawhney (NVIDIA); Daqi Lin (NVIDIA, USA); Markus Kettunen (NVIDIA, Finnl and); Benedikt Bitterli (NVIDIA, USA); Ravi Ramamoorthi (University of Cal ifornia San Diego; NVIDIA, USA); Chris Wyman and Matt Pharr (NVIDIA, USA); and Rohan Sawhney\n---------------------\nAONeuS: A Neural Rendering Fram ework for Acoustic-optical Sensor Fusion\n\nUnderwater 3D surface reconstr uction has broad applications. Often, submersibles must capture measuremen ts from a small baseline. Small baselines make 3D reconstruction more chal lenging. We present an acoustic-optical neural surface reconstruction fram ework (AONeuS), which fuses complementary sonar ...\n\n\nMohamad Qadri (Ca rnegie Mellon University), Kevin Zhang (University of Maryland College Par k), Akshay Hinduja and Michael Kaess (Carnegie Mellon University), Adithya Pediredla (Dartmouth College), and Christopher Metzler (University of Mar yland College Park)\n---------------------\nAnalogist: Out-of-the-box Visu al In-context Learning With Image Diffusion Model\n\nAnalogist is a novel visual In-context Learning approach combining visual and textual prompts w ith a diffusion model. It introduces self-attention cloning and cross-atte ntion masking to enhance analogy accuracy, offering a flexible, out-of-the -box solution that outperforms existing methods without n...\n\n\nZheng Gu (Nanjing University, City University of Hong Kong); Shiyuan Yang (Tianjin University, City University of Hong Kong); Jing Liao (City University of Hong Kong); and Jing Huo and Yang Gao (Nanjing University)\n-------------- -------\nImportance Sampling BRDF Derivatives\n\nWe propose the first set of techniques to efficiently importance sample the derivatives of several popular analytical BRDF models (e.g. anisotropic GGX/ Beckmann/ Ashikhmin- Shirley, Hanrahan-Kreuger, Oren-Nayar etc.). Our estimators are practical and provide significant variance improvement in all s...\n\n\nYash Belhe a nd Bing Xu (University of California San Diego), Sai Praveen Bangaru (Mass achusetts Institute of Technology (MIT)), Ravi Ramamoorthi and Tzu-Mao Li (University of California San Diego), and Yash Belhe\n-------------------- -\nStopThePop: Sorted Gaussian Splatting for View-consistent Real-time Ren dering\n\nStopThePop, our novel hierarchical rasterizer, enhances 3D Gauss ian Splatting by eliminating popping artifacts caused by global sorting. O ur proposed hierarchical, approximate per-pixel sort significantly improve s view-consistency for novel-view synthesis, while sophisticated culling a nd load balanc...\n\n\nLukas Radl and Michael Steiner (Graz University of Technology); Mathias Parger (Independent); Alexander Weinrauch (Graz Unive rsity of Technology); Bernhard Kerbl (Technische Universität Wien (TU Wien )); and Markus Steinberger (Graz University of Technology, Huawei Technolo gies)\n---------------------\nModal Folding: Discovering Smooth Folding Pa tterns for Sheet Materials Using Strain-space Modes\n\nIn this paper, we d evelop a computational approach for automatically discovering complex fold ing patterns. We demonstrate the effectiveness of our method with simulati on results for a range of shapes and materials, and we validate our design s with physical prototypes.\n\n\nPengbin Tang (ETH Zürich, Université de M ontréal); Ronan Hinchet (ETH Zürich); Roi Poranne (University of Haifa, ET H Zürich); and Bernhard Thomaszewski and Stelian Coros (ETH Zürich)\n----- ----------------\nIn the Quest for Scale-Optimal Mappings\n\nWe present an algorithm for the max-norm minimization of hyperelastic distortion for me sh deformation.\nUnder certain conditions our method can build a deformati on whose maximum distortion is arbitrarily close to the (unknown) minimum. \nIn summary, we reliably build 2D and 3D mesh deformations with th...\n\n \nDmitry Sokolov (Université de Lorraine, CNRS, Inria, LORIA); Vladimir Ga ranzha, Igor Kaporin, and Liudmila Kudryavtseva (Dorodnicyn Computing Cent er FRC CSC RAS Moscow Institute of Physics and Technology); François Prota is (Université de Lorraine, CNRS, Inria, LORIA); and Dmitry Sokolov\n----- ----------------\nMERCI: Mixed curvature-based elements for computing equi libria of thin elastic ribbons\n\nRibbons are thin elastic structures lyin g in between rods and plates. Relying on the 1D Wunderlich ribbon model, w e propose mixed position-curvature elements for computing ribbon equilibri a subject to arbitrary boundary constraints. Our simulator is carefully va lidated on subtle scenarios such as Mö...\n\n\nRaphaël Charrondière (Unive rsity Grenoble Alpes Inria, CNRS, Grenoble INP, LJK); Sébastien Neukirch ( Sorbonne Université - CNRS); Florence Bertails-Descoubes (University Greno ble Alpes Inria, CNRS, Grenoble INP, LJK); and Sébastien Neukirch and Flor ence Bertails-Descoubes\n---------------------\nModeling Hair Strands With Roving Capsules\n\nThe paper derives a parametric representation for swep t sphere shapes and proposes an efficient ray-surface intersection algorit hm using capsules that are dynamically defined per ray at runtime. This ne w algorithm is more than twice as fast as previous approaches and can be u sed for hair and fur ren...\n\n\nAlexander Reshetov and David Hart (NVIDIA )\n---------------------\nDeep Sketch Vectorization via Implicit Surface E xtraction\n\nWe propose a sketch vectorization algorithm based on surface extraction from unsigned distance fields. Our representation is naturally controllable, which we demonstrate in an interactive topology refinement i nterface. We achieve far more accurate vectorizations on complex input tha n previous approa...\n\n\nChuan Yan (George Mason University); Yong Li (So uth China University of Technology, George Mason University); Deepali Anej a and Matthew Fisher (Adobe); Edgar Simo-Serra (Waseda University); and Yo tam Gingold (George Mason University)\n---------------------\nBoostMVSNeRF s: Boosting MVS-based NeRFs to Generalizable View Synthesis in Large-scale Scenes\n\nBoostMVSNeRFs revolutionizes 3D scene visualization by improvin g the quality of large-scale MVS-based NeRFs. This innovative technique co mbines multiple cost volumes with 3D visibility scores to enhance novel vi ew synthesis without additional training. Compatible with current framewor ks, it sets new...\n\n\nChih-Hai Su (National Yang Ming Chiao Tung Univers ity); Chih-Yao Hu (National Taiwan University); and Shr-Ruei Tsai, Jie-Yin g Lee, Chin-Yang Lin, and Yu-Lun Liu (National Yang Ming Chiao Tung Univer sity)\n---------------------\nSMERF: Streamable Memory Efficient Radiance Fields for Real-time Large-scene Exploration\n\nSMERF is a method for real -time view-synthesis of large, multi-room spaces on resource-constrained d evices. By distilling state-of-the-art NeRF models into a compact, hierarc hical representation, SMERF enables the exploration of spaces as large as 300 m^2 with quality approaching that of its teacher...\n\n\nDaniel Duckwo rth (Google DeepMind); Peter Hedman (Google Research); Christian Reiser (G oogle Research; Tübingen AI Center, University of Tübingen); Peter Zhizhin (Google Research); Jean-François Thibert (Google AR/VR); Mario Lučić (Goo gle DeepMind); and Richard Szeliski and Jonathan T. Barron (Google Researc h)\n---------------------\nUniversal Facial Encoding of Codec Avatars From VR Headsets\n\nWe present a self-supervised and robust encoding algorithm for achieving high-fidelity real-time 3D facial animation, via head-mount ed cameras on a consumer VR headset. Our model generalizes to unseen users and variability in illuminations, under incomplete views of the face, ena bling accessible aut...\n\n\nShaojie Bai, Te-Li Wang, Chenghui Li, Akshay Venkatesh, Tomas Simon, Chen Cao, Gabriel Schwartz, Jason Saragih, Yaser S heikh, and Shih-En Wei (Facebook Reality Labs)\n---------------------\nNeu ralTO: Neural Reconstruction and View Synthesis of Translucent Objects\n\n We introduce a novel, two-stages framework, which is geared toward high-fi delity surface reconstruction and the novel-view synthesis of translucent objects. In our framework, we propose a theoretical model for the neural r adiance field of translucent objects, which parametrizes the density field usi...\n\n\nYuxiang Cai and Jiaxiong Qiu (Nankai University, TMCC); Zhong Li (OPPO US Research Center); and Bo Ren (Nankai University, TMCC)\n----- ----------------\nRay Tracing Harmonic Functions\n\nWe introduce a sphere- tracing-like method for visualizing surfaces encoded as level sets of harm onic functions. We show examples visualizing smooth surfaces from point cl ouds or polygon soups (without linear solves or mesh extraction), nonplana r polygons, architectural grid shells, mesh "exoskeleton...\n\n\nMark Gill espie (Carnegie Mellon University); Denise Yang (Pixar, Carnegie Mellon Un iversity); Mario Botsch (Technical University of Dortmund); and Keenan Cra ne (Carnegie Mellon University)\n---------------------\nSolid Knitting\n\n We introduce solid knitting, a new fabrication technique combining the lay er-by-layer volumetric approach of 3D printing with the topologically entw ined stitch structure of knitting to produce solid 3D objects. We present a solid knitting machine, a design tool using a volumetric block structure , an...\n\n\nYuichi Hirose and Mark Gillespie (Carnegie Mellon University) ; Angelica M. Bonilla Fominaya (Carnegie Mellon University, Google); and J ames McCann (Carnegie Mellon University)\n---------------------\nMVD^2: Ef ficient Multiview 3D Reconstruction for Multiview Diffusion\n\nMultiview d iffusion (MVD) has emerged as a prominent 3D generation technique, but fac es challenges with inconsistency and view sparseness, impacting the qualit y of multiview 3D reconstruction. Our learning-based MVD^2 method tackles these challenges, ensuring efficient and robust 3D reconstruction w...\n\n \nXin-Yang Zheng (Tsinghua University, Microsoft Research Asia); Hao Pan, Yu-Xiao Guo, and Xin Tong (Microsoft Research Asia); and Yang Liu (Microso ft)\n---------------------\nRepulsive Shells\n\nShape spaces are a powerfu l tool for nonlinear interpolation, extrapolation, and averaging of geomet ric data, but previous shape spaces permit geometry to self-intersect in n onphysical ways. We introduce a shape space where geometry naturally avoid s intersection, as well as an adaptive collision pot...\n\n\nJosua Sassen (University of Bonn, École Normale Supérieure Paris-Saclay); Henrik Schuma cher (University of Georgia); Martin Rumpf (University of Bonn); and Keena n Crane (Carnegie Mellon University)\n---------------------\nComputational Illusion Knitting\n\nIllusion-knit fabrics reveal hidden images across vi ewing angles. Artist-created knit illusions are tedious to design, limited to single-view, and slow to manufacture. We establish constraints over th e design space, develop an interactive design system, and originate fabric ation techniques for mixed...\n\n\nAmy Zhu, Yuxuan Mei, Benjamin Jones, Za chary Tatlock, and Adriana Schulz (University of Washington)\n------------ ---------\nFabric Tessellation: Realizing Freeform Surfaces by Smocking\n\ nThe paper introduces a method for utilizing an embroidery technique calle d smocking to realize freeform surfaces from a flat piece of fabric. By co mbining directional field computation and continuous planar graph optimiza tion, our algorithm outputs a stitching pattern that generates an approxim ation...\n\n\nAviv Segall and Jing Ren (ETH Zürich), Amir Vaxman (Universi ty of Edinburgh), and Olga Sorkine-Hornung (ETH Zürich)\n----------------- ----\nWalkTheDog: Cross-morphology Motion Alignment via Phase Manifolds\n\ nWe present a morphology and skeletal structure-independent approach for u nderstanding the periodicity structure and semantics of motion datasets. D riven by the common phase manifold with multiple closed curves learned wit h vector-quantized periodic autoencoders, a precise semantic and timing al ignme...\n\n\nPeizhuo Li (ETH Zürich), Sebastian Starke and Yuting Ye (Met a Reality Labs), and Olga Sorkine-Hornung (ETH Zürich)\n------------------ ---\nControllable Neural Style Transfer for Dynamic Meshes\n\nA novel mesh neural style transfer technique is presented. We improve upon the previou s stylization works through implicit re-parametrizations of meshes, contro llable style orientations, better temporal coherency treatment, and volume conservation. These improvements enable high-quality mesh styliza...\n\n\ nGuilherme Gomes Haetinger, Jingwei Tang, and Raphael Ortiz (Disney Resear ch Studios); Paul Kanyuk (Pixar Animation Studios); and Vinicius Azevedo ( Disney Research Studios)\n\nInterest Area: Arts & Design, Production & Ani mation, Research & Education\n\nRecording: Livestreamed, Recorded\n\nRegis tration Category: Full Conference, Full Conference Supporter, Virtual Acce ss, Experience, Exhibitor Full Conference, Exhibitor Experience, Sunday END:VEVENT END:VCALENDAR