SIGGRAPH 2025 Papers with AI Summaries
ACM SIGGRAPH (Transactions on Graphics) · 306 papers
→ SIGGRAPH 2025 papers with code (46)
Each paper below shows an AI-generated one-line summary. Get the full 6-part summary (innovation, method, data, results, limitations) and search all 306 SIGGRAPH 2025 papers by keyword, author or institution —
free trial on arXivSub.
3D Stylization via Large Reconstruction Model
Ipek Oztas (Bilkent University), Aysegul Dundar (Bilkent University)
Image TranslationGenerationOptimizationTransformerDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningGaussian SplattingImagePoint CloudMesh
🎯 What it does: Inject the visual style of reference images into 3D geometry using large 3D reconstruction models (e.g., InstantMesh), achieving stylization of 3D objects without the need for training or testing during optimization.
3D-Fixup: Advancing Photo Editing with 3D Priors
Yen-Chi Cheng (University of Illinois Urbana-Champaign), Nanxuan Zhao (Adobe Research)
Image TranslationImage HarmonizationRestorationGenerationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderOptical FlowImageVideoPoint CloudMesh
🎯 What it does: This paper proposes 3D-Fixup, a diffusion model guided by optical flow based on 3D priors, which can perform large-scale 3D translation and rotation edits on a single image while maintaining identity consistency.
3DGH: 3D Head Generation with Composable Hair and Face
Chengan He (Yale University), Giljoo Nam (Meta Codec Avatars Lab)
GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingImageMesh
🎯 What it does: Propose the 3DGH model to achieve fully viewable renderable 3D faces and composable hairstyle generation.
3DGS2: Near Second-order Converging 3D Gaussian Splatting
Lei Lan (University of Utah), Yin Yang (University of Utah)
GenerationData SynthesisOptimizationComputational EfficiencyLarge Language ModelDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingImagePoint CloudBenchmark
🎯 What it does: Propose a 3D Gaussian Splatting (3DGS) training algorithm based on local Newton optimization, significantly improving convergence speed while maintaining or enhancing reconstruction quality.
4D Gaussian Videos with Motion Layering
Pinxuan Dai, Weiwei Xu
CompressionNeural Radiance FieldGaussian SplattingOptical FlowVideo
🎯 What it does: Propose a 4D Gaussian Video (4DGV) method for creating and streaming realistic volumetric videos of dynamic scenes.
A Deep Learning-based Virtual Oculoplastic Surgery Simulator
Seonghyeon Kim, Jun-yong Noh
Image TranslationGenerationData SynthesisPrompt EngineeringDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkImageMesh
🎯 What it does: Proposed a virtual orbital surgery simulation system based on deep learning, aiming to reduce patients' anxiety about surgical outcomes through realistic previews of surgical results and to facilitate communication between doctors and patients.
A Divide-and-Conquer Approach for Global Orientation of Non-Watertight Scene-Level Point Clouds with 0-1 Integer Optimization
Zhuodong Li (Chinese Academy of Sciences), Ying He (Nanyang Technological University)
OptimizationPoint Cloud
🎯 What it does: This paper proposes the DACPO (Divide-and-Conquer Point Orientation) framework for achieving global normal orientation in non-closed, scene-level point clouds. First, the point cloud is divided into several mutually connected small blocks, and each block is independently processed with normal initialization and iterative Poisson reconstruction (iPSR) to obtain locally consistent normals. Subsequently, an inter-block graph is built based on visible connected regions (VCR), and a 0-1 integer optimization is applied to determine whether each block needs to be flipped, thereby achieving globally consistent orientation across the entire scene.
A Fast Parallel Median Filtering Algorithm Using Hierarchical Tiling
Louis Sugy (NVIDIA)
RestorationComputational EfficiencyDiffusion modelAuto EncoderContrastive LearningOptical FlowImage
🎯 What it does: A fast parallel median filtering algorithm based on hierarchical partitioning is proposed, and two variants, data-independent and data-dependent, are implemented.
A Fluorescent Material Model for Non-Spectral Editing & Rendering
Laurent Belcour (Intel Corporation), Pascal Barla (Inria Bordeaux)
Contrastive LearningGaussian SplattingImageTabularBenchmarkPhysics Related
🎯 What it does: Propose a fluorescent material model suitable for non-spectral renderers, using normalized two-dimensional Gaussian mixtures to approximate fluorescent re-emission, and enabling control of fluorescent parameters during real-time editing.
A Fully-statistical Wave Scattering Model for Heterogeneous Surfaces
Zhengze Liu, Rui Wang
Physics Related
🎯 What it does: A model is proposed that provides a fully statistical description of heterogeneous surfaces at the microscale, and calculates their BRDF and scattering field covariance based on the Harvey-Shack theory, thereby enabling sampling of light spots and enriching surface appearance without explicitly defining the surface.
A Hybrid Near-wall Model for Kinetic Simulation of Turbulent Boundary Layer Flows
Mengyun Liu, Xiaopei Liu
Diffusion modelScore-based ModelFlow-based ModelRectified FlowNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingPhysics RelatedStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A hybrid wall model is proposed, combining macroscopic and mesoscopic algebraic models, applied to a lattice Boltzmann solver to handle turbulent boundary layers, utilizing multi-resolution techniques and parameterization to adapt to different solid surface properties.
A Monte Carlo Rendering Framework for Simulating Optical Heterodyne Detection
Juhyeon Kim, A. Pediredla
Physics Related
🎯 What it does: Propose a flexible and general spectral-domain optical stray interference detection (OHD) simulation framework based on classical radiometric path integration and combined with Monte Carlo path tracing technology;
A Platform for Interactive AI Character Experiences
Rafael Wampfler (ETH Zurich), Markus Gross (ETH Zurich)
GenerationData SynthesisRecommendation SystemExplainability and InterpretabilityComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelPrompt EngineeringVision-Language-Action ModelDiffusion modelAuto EncoderGenerative Adversarial NetworkVideoTextMultimodalityRetrieval-Augmented GenerationAudio
🎯 What it does: Built an scalable digital character interaction platform, demonstrating dialogues and narrative-driven experiences with a virtual Einstein avatar.
A Polyhedral Construction of Empty Spheres in Discrete Distance Fields
M. Kohlbrenner, Marc Alexa
Optimization
🎯 What it does: A method for constructing polyhedra based on Lie spherical geometry is proposed, which identifies empty balls (balls that do not intersect with a given set of balls) and uses them for isosurface sampling of discrete signed distance fields; by representing the space of non-intersecting balls as the intersection of half-spaces and tetrahedra, the intersection points between the boundary of the maximum ball and the tetrahedron are computed;
A Versatile Quaternion-Based Constrained Rigid Body Dynamics
Guirec Maloisel, Moritz Bächer
Physics RelatedOrdinary Differential Equation
🎯 What it does: Proposes a rigid body dynamics method that satisfies constraints through an implicit integration scheme and employs additive quaternion updates, supporting complex mechanical systems with arbitrary kinematic structures;
Accelerated Gamut Discovery via Massive Parallelization
Navid Ansari, Vahid Babaei
OptimizationHyperparameter Search
🎯 What it does: Propose an scalable framework for efficiently discovering performance domains of different processes.
Adaptive Algebraic Reuse of Reordering in Cholesky Factorizations with Dynamic Sparsity Patterns
Behrooz Zarebavani (University of Toronto), Maryam Mehri Dehnavi (NVIDIA)
OptimizationComputational EfficiencyGraphPhysics Related
🎯 What it does: Propose Parth, a framework that can adaptively reuse the fill-reducing ordering in sparse Cholesky solvers;
Adaptive Phase-Field-FLIP for Very Large Scale Two-Phase Fluid Simulation
Bernhard Braun, N. Thuerey
Optical FlowPhysics Related
🎯 What it does: Proposes the PF-FLIP hybrid Eulerian/Lagrangian method, achieving physically based simulations for large-scale, high Reynolds number, high density ratio two-phase flows using spatially adaptive and dual-resolution schemes, supporting billions of particles and thousands of grid resolution levels.
Aerial Path Online Planning for Urban Scene Updation
Mingfeng Tang (Shenzhen University), Hui Huang (Shenzhen University)
Autonomous DrivingOptimizationComputational EfficiencyPrompt EngineeringNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingOptical FlowImageVideoPoint CloudMeshRetrieval-Augmented Generation
🎯 What it does: Propose a two-step UAV path planning method based on prior reconstruction models and change probability statistics, specifically designed for change detection and local updates in urban scenarios.
AlignTex: Pixel-Precise Texture Generation from Multi-view Artwork
Yuqing Zhang, Xiaogang Jin
GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldGenerative Adversarial NetworkImageMesh
🎯 What it does: Propose the AlignTex framework, which generates high-quality textures by utilizing multi-view artworks and 3D meshes, and achieves pixel-level details and geometric consistency through a two-stage process (image alignment generation and texture refinement)
AMOR: Adaptive Character Control through Multi-Objective Reinforcement Learning
L. N. Alegre, Moritz Bacher
Robotic IntelligenceReinforcement LearningGenerative Adversarial NetworkContrastive LearningVideoTabularTime SeriesSequential
🎯 What it does: This paper proposes AMOR, a physics-based character control framework based on multi-objective reinforcement learning. The trained policies can dynamically adjust according to reward weights immediately after training, and through a hierarchical high-level policy, it further enables real-time weight selection of implicit rewards, thus achieving efficient control in multi-poses, dynamic, and robotic scenarios.
ANIME-Rod: Adjustable Nonlinear Isotropic Materials for Elastic Rods
Huanyu Chen, J. Barbič
Physics Related
🎯 What it does: Propose a 3D elastic rod large deformation simulation method based on an arbitrary nonlinear isotropic 3D solid elastic energy density function ψ, deriving the rod's elastic energy through the limit process of the rod thickness approaching zero, and uniformly describing the rod's extension, bending, and torsion.
Anymate: A Dataset and Baselines for Learning 3D Object Rigging
Yufan Deng (Stanford University), Jiajun Wu (Stanford University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelAuto EncoderContrastive LearningPoint CloudMeshGraphBenchmark
🎯 What it does: This paper proposes the Anymate dataset and a three-stage learning framework for automatically generating skeletal assemblies and skin weights from static 3D meshes, achieving fully automatic 3D object animation.
AnyTop: Character Animation Diffusion with Any Topology
Inbar Gat (Tel Aviv University), Daniel Cohen-Or (Tel Aviv University)
GenerationPose EstimationTransformerPrompt EngineeringDiffusion modelTextGraph
🎯 What it does: Proposes AnyTop, a general-purpose character animation generation framework based on diffusion models, capable of generating diverse and natural motion sequences using arbitrary skeletal topology information.
Appearance-aware Multi-view SVBRDF Reconstruction via Deep Reinforcement Learning
Pengfei Zhu, Yanwen Guo
Reinforcement Learning
🎯 What it does: Designed and implemented an adaptive sampling method based on deep reinforcement learning to determine the optimal view-lighting combination for reconstructing spatially varying bidirectional reflectance distribution functions (SVBRDF)
Arenite: A Physics-based Sandstone Simulator
Zhanyu Yang, Bedrich Benes
Physics Related
🎯 What it does: Introduces a physics-based sandstone simulator called Arenite, which simulates the combination of stress and multi-factor erosion to generate various sandstone structures found in nature.
AssetDropper: Asset Extraction via Diffusion Models with Reward-Driven Optimization
Lanjiong Li (Hong Kong University of Science and Technology), Zeyu Wang (Hong Kong University of Science and Technology)
GenerationData SynthesisOptimizationConvolutional Neural NetworkTransformerReinforcement LearningPrompt EngineeringVision Language ModelDiffusion modelScore-based ModelImageTextMultimodality
🎯 What it does: Proposed a diffusion model-based asset extraction framework called AssetDropper, which can accurately extract standardized assets from any image region and directly use them for subsequent design tasks.
Asymptotic analysis and design of linear elastic shell lattice metamaterials
Di Zhang (University of Science and Technology of China), Ligang Liu (University of Science and Technology of China)
OptimizationMeshPhysics Related
🎯 What it does: Based on Ciarlet's thin shell theory, an asymptotic analysis is conducted on shell lattice structures with thickness approaching zero, introducing and strictly defining 'Asymptotic Directional Stiffness (ADS)', and providing its convergence theorem, upper bound, and optimality conditions; subsequently, a discrete and shape optimization framework based on triangular mesh is developed, enabling efficient design of periodic mid-surface structures such as TPMS.
Augmented Vertex Block Descent
Chris Giles, C. Yuksel
OptimizationPhysics Related
🎯 What it does: Extend Vertex Block Descent by adopting an augmented Lagrangian formulation to address convergence issues in hard constraints and high stiffness ratios, supporting complex contact scenarios such as stacked friction and finite-degree-of-freedom joints in physical simulations.
AutoKeyframe: Autoregressive Keyframe Generation for Human Motion Synthesis and Editing
Bowen Zheng, Xiaogang Jin
GenerationPose EstimationTransformerDiffusion modelScore-based ModelSequential
🎯 What it does: Propose the AutoKeyframe framework, which directly generates keyframes to simultaneously accept dense and sparse control signals, achieving the synthesis and editing of human motion;
Automated Task Scheduling for Cloth and Deformable Body Simulations in Heterogeneous Computing Environments
Chen He, Huamin Wang
OptimizationComputational EfficiencyDiffusion modelGaussian SplattingOptical FlowMeshPhysics Related
🎯 What it does: Developed an automatic task scheduling framework to optimize the performance of fabric and deformable body simulations in heterogeneous computing environments.
BANG: Dividing 3D Assets via Generative Exploded Dynamics
Longwen Zhang (ShanghaiTech University), Jingyi Yu (ShanghaiTech University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningVideoPoint CloudMesh
🎯 What it does: BANG achieves decomposable and reconfigurable 3D assets by generating continuous explosion dynamics, supporting component-level refinement, controllable explosions, and post-processing detail enhancement.
Be Decisive: Noise-Induced Layouts for Multi-Subject Generation
Omer Dahary, Daniel Cohen-Or
GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelScore-based ModelAuto EncoderContrastive LearningImageText
🎯 What it does: Propose a multi-agent generation method based on noise-induced layout, automatically predicting and iteratively optimizing the layout to control the generation of multi-agent images.
Bernstein Bounds for Caustics
Zhimin Fan (Nanjing University), Jie Guo (Nanjing University)
Diffusion modelNeural Radiance FieldOptical FlowMeshPhysics Related
🎯 What it does: Propose a method based on Bernstein polynomials for position information and radiance upper bounds, performing beam tracing and partitioning of triangle tuples during the precomputation phase, and subsequently using these upper bounds for random sampling during rendering, significantly reducing the variance of high-frequency caustics.
Boolean Operation for CAD Models Using a Hybrid Representation
Yingyu Yang, Dong-ming Yan
Mesh
🎯 What it does: An algorithm is proposed to achieve efficient and accurate Boolean operations by mapping B-Rep models to triangular meshes with controllable error.
BrepDiff: Single-Stage B-rep Diffusion Model
Mingi Lee, Young Min Kim
GenerationData SynthesisTransformerDiffusion modelMesh
🎯 What it does: Propose a single-stage B-rep diffusion model called BrepDiff for generating 3D B-rep structures;
BuildingBlock: A Hybrid Approach for Structured Building Generation
Junming Huang (Zhejiang University), Weiwei Xu (Zhejiang University)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringDiffusion modelGenerative Adversarial NetworkImageTextMesh
🎯 What it does: Propose a two-stage hybrid method called BuildingBlock, combining generative models, LLM, and PCG, to generate structured building models from text prompts and support local editing.
C-Tubes: Design and Optimization of Tubular Structures Composed of Developable Strips
Michele Vidulis, Mark Pauly
Optimization
🎯 What it does: Proposed the concept of C-tubes and designed construction algorithms and form discovery tools to ensure developable surfaces for generating 3D tubular structures.
C5D: Sequential Continuous Convex Collision Detection Using Cone Casting
Xiao Yuan, Hao Su
OptimizationComputational EfficiencyOptical FlowSequential
🎯 What it does: Propose a sequential continuous convex collision detection (CCD) algorithm for collision detection of convex objects under constant affine motion, utilizing the conservative advancement method to iteratively refine the time of impact (TOI) estimation, and seamlessly integrate with the ABD framework.
CageNet: A Meta-Framework for Learning on Wild Meshes
Michal Edelstein (Technion Israel Institute of Technology), M. Ben-Chen
Representation LearningMeta LearningGraph Neural NetworkDiffusion modelAuto EncoderContrastive LearningMesh
🎯 What it does: Propose the CageNet meta-framework, which encloses any wild mesh (multi-connected, non-manifold, etc.) with a single manifold cage, and uses generalized Basel coordinates to map the learned features from the cage back to the original mesh, enabling seamless learning and inference on wild meshes.
CAST: Component-Aligned 3D Scene Reconstruction from an RGB Image
Kaixin Yao (ShanghaiTech University), Jingyi Yu (ShanghaiTech University)
SegmentationGenerationPose EstimationDepth EstimationOptimizationTransformerPrompt EngineeringVision Language ModelDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkImageTextPoint CloudMeshRetrieval-Augmented Generation
🎯 What it does: Proposes CAST, a method for component-aligned 3D scene reconstruction from a single RGB image, capable of generating high-quality meshes, textures, and maintaining physical consistency.
CineMaster: A 3D-Aware and Controllable Framework for Cinematic Text-to-Video Generation
Qinghe Wang (Dalian University of Technology), Kun Gai (Kuaishou Technology)
GenerationPose EstimationDepth EstimationTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningGaussian SplattingImageVideoText
🎯 What it does: Propose the CineMaster framework, which enables interactive layout of objects and cameras in 3D space, and generates 3D-aware, controllable text-to-video models based on this.
Cirrus: Adaptive Hybrid Particle-Grid Flow Maps on GPU
Mengdi Wang, Bo Zhu
OptimizationComputational EfficiencyFlow-based ModelOptical FlowPhysics Related
🎯 What it does: Proposed and implemented an adaptive hybrid particle-grid flow map method, constructing a fully adaptive flow map-based fluid simulation framework.
CK-MPM: A Compact-Kernel Material Point Method
Michael Liu (Carnegie Mellon University), Minchen Li (Carnegie Mellon University)
Diffusion modelOptical FlowPoint CloudPhysics RelatedStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose a C² continuous, radius-2 compact kernel Material Point Method (CK-MPM), and achieve robust mapping between particles and grids through a dual-grid framework, compatible with APIC and MLS-MPM;
Claycode: Stylable and Deformable 2D Scannable Codes
Marco Maida (Independent Researcher), Elena Camuffo (Independent Researcher)
RecognitionDiffusion modelContrastive LearningOptical FlowImageVideoMesh
🎯 What it does: The paper proposes a highly stylizable and deformable 2D scannable code called Claycode.
Clebsch Gauge Fluid on Particle Flow Maps
Zhiqi Li, Bo Zhu
Optical FlowPhysics Related
🎯 What it does: Proposed a novel gauge fluid solver that evolves the Clebsch wave function on a particle flow map.
Closed-form Generalized Winding Numbers of Rational Parametric Curves for Robust Containment Queries
Shibo Liu, Xiao-Ming Fu
Symbolic ComputationOptimizationComputational EfficiencyScore-based Model
🎯 What it does: A closed-form expression for the generalized winding number of rational parametric curves is derived, along with a method for computing its derivatives, enabling robust enclosure queries.
CLR-Wire: Towards Continuous Latent Representations for 3D Curve Wireframe Generation
Xueqi Ma (Shenzhen University), Hui Huang (Shenzhen University)
GenerationRepresentation LearningTransformerDiffusion modelFlow-based ModelAuto EncoderPoint CloudMeshOrdinary Differential Equation
🎯 What it does: Propose the CLR-Wire framework, which uniformly encodes the geometry and topology of 3D curve wireframes into a continuous fixed-length latent space, and achieves unconditional and conditional generation through a flow matching model;
CMD: Controllable Multiview Diffusion for 3D Editing and Progressive Generation
Peng Li (Hong Kong University of Science and Technology), Yi-Ting Guo
GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderImagePoint CloudMesh
🎯 What it does: Proposed the CMD framework, which achieves local editing of 3D models (only modifying one rendered image) and the ability to progressively generate complex 3D models through a conditional multi-view diffusion model.
Cobra: Efficient Line Art COlorization with BRoAder References
Junhao Zhuang, Ying Shan
Image TranslationGenerationComputational EfficiencyTransformerPrompt EngineeringDiffusion modelScore-based ModelImageRetrieval-Augmented Generation
🎯 What it does: Proposed a long-context reference line drawing coloring framework called Cobra, which can efficiently and accurately color comic line drawings and supports more than 200 reference images and color prompts.
Collaborative On-Sensor Array Cameras
Jipeng Sun (Princeton University), Felix Heide (Princeton University)
RestorationOptimizationDiffusion modelAuto EncoderContrastive LearningOptical FlowImageMultimodalityPhysics Related
🎯 What it does: This paper proposes a collaborative photodetector array camera using a 2×3 metal lens array (comprising one million nanocolumns), and achieves high-quality wide-field imaging across the entire visible light spectrum through end-to-end differentiable optical modeling and image reconstruction network joint optimization.
Color Matching and Biomimicry for Multi-Material Dental 3D Printing
András Simon, Henning Lübbe
Diffusion modelScore-based ModelAuto EncoderGenerative Adversarial NetworkMesh
🎯 What it does: Proposing a multi-material 3D printing method for dental hierarchical biomimicry and multi-point color difference matching
ColorSurge: Bringing Vibrancy and Efficiency to Automatic Video Colorization via Dual-Branch Fusion
Hongbo Zhao, Yijun Wang
Image TranslationRestorationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderContrastive LearningVideo
🎯 What it does: Propose a lightweight end-to-end video colorization network called ColorSurge
Compensating Spatiotemporally Inconsistent Observations for Online Dynamic 3D Gaussian Splatting
Youngsik Yun (Yonsei University), Youngjung Uh (Yonsei University)
OptimizationComputational EfficiencyNeural Radiance FieldAuto EncoderGaussian SplattingImageVideoPoint Cloud
🎯 What it does: Propose an online dynamic 3D Gaussian splatting method that significantly improves spatiotemporal consistency by learning residual maps to recover an ideal scene from noisy observations.
Computational Modeling of Gothic Microarchitecture
Aviv Segall, O. Sorkine-Hornung
RestorationGenerationDiffusion modelAuto EncoderGenerative Adversarial NetworkImageMeshReview/Survey Paper
🎯 What it does: Propose a framework based on simple geometric principles, utilizing historical two-dimensional drawings to model Gothic miniature architecture as a three-dimensional curve network;
Conformal First Passage for Epsilon-free Walk-on-Spheres
Paul Himmler, Tobias Günther
Physics RelatedStochastic Differential Equation
🎯 What it does: Propose a method that uses geometric primitives and employs conformal mapping to solve the two-dimensional Laplace equation, avoiding the epsilon-shell error in the traditional Walk-on-Spheres method.
Controllable Complex Freezing Dynamics Simulation on Thin Films
Yijie Liu, Bo-Ning Ren
Diffusion modelOptical FlowMeshPhysics Related
🎯 What it does: Physically simulate the freezing dynamics on thin films and realize controllable dendritic crystal freezing patterns on the MELP framework using the Phase Map method.
Controllable Tracking-Based Video Frame Interpolation
Karlis Martins Briedis, Christopher Schroers
Object TrackingGenerationPrompt EngineeringDiffusion modelOptical FlowVideo
🎯 What it does: Propose a video frame interpolation method based on sparse point tracking, and provide user-interactable tracking refinement and model weight adaptation during inference to control the level of hallucination.
Cora: Correspondence-aware image editing using few step diffusion
Amirhossein Alimohammadi (Simon Fraser University), Ali Mahdavi-Amiri (Simon Fraser University)
Image TranslationImage HarmonizationGenerationTransformerPrompt EngineeringDiffusion modelScore-based ModelImage
🎯 What it does: Proposed a image editing framework called Cora based on a few-step diffusion model, which can achieve various editing tasks such as non-rigid deformation, object insertion/deletion, and texture modification.
Correct your balance heuristic: Optimizing balance-style multiple importance sampling weights
Qingqin Hua, Philipp Slusallek
Optimization
🎯 What it does: A general and practical weight correction scheme is proposed, which improves the weight function of MIS by online optimizing a set of correction factors and multiplying them into any baseline MIS heuristic (such as balance or power).
Creating Fluid-Interactive Virtual Agents by an Efficient Simulator with Local-domain Control
Wenbin Song, Xiaopei Liu
Computational EfficiencyRobotic IntelligencePhysics Related
🎯 What it does: A fluid-structure coupling simulator based on an efficient Lattice Boltzmann solver was developed, which uses dynamically moving local domains enclosing virtual agents to achieve more flexible and efficient control strategy acquisition;
CueTip: An Interactive and Explainable Physics-aware Pool Assistant
Sean Memery (University of Edinburgh), Kartic Subr (University of Edinburgh)
Explainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningTextPhysics RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Built an interactive, interpretable automated billiards coaching assistant called CueTip, which provides users with shooting advice and explanations tailored to specific table states by combining natural language interfaces, physics simulations, and rule explanations.
Curl Quantization for Automatic Placement of Knit Singularities
Rahul Mitra, Edward Chien
OptimizationDiffusion modelOptical FlowMesh
🎯 What it does: Developed an automatic knitted singularity placement method based on curl quantization, generating stripe patterns that match the contour lines of the knitting time function.
DAM-VSR: Disentanglement of Appearance and Motion for Video Super-Resolution
Zhe Kong (Sun Yat-sen University), Wenhan Luo (HKUST)
Super ResolutionTransformerDiffusion modelAuto EncoderOptical FlowVideo
🎯 What it does: Proposes a decoupled framework called DAM-VSR, which decomposes video super-resolution into appearance enhancement and motion control, in order to fully utilize the detail generation capability of image super-resolution models and the generative prior of video diffusion models.
Data-Efficient Discovery of Hyperelastic TPMS Metamaterials with Extreme Energy Dissipation
Maxine Perroni-Scharf (Massachusetts Institute of Technology), Mina Konaković Luković (Massachusetts Institute of Technology)
OptimizationData-Centric LearningDrug DiscoveryConvolutional Neural NetworkRecurrent Neural NetworkTransformerDiffusion modelScore-based ModelFlow-based ModelPoint CloudMeshGraphTabularAlzheimer's DiseaseElectronic Health RecordsReview/Survey PaperBenchmarkPhysics Related
🎯 What it does: By performing implicit interpolation on eight existing triply periodic minimal surface (TPMS) prototypes to construct a broad design space, and then using Bayesian optimization and deep ensemble models to experimentally screen micro-scale super-elastic TPMS structures within this space, the study ultimately discovers and verifies structures with energy dissipation rates twice as high as existing prototypes.
DC-VSR: Spatially and Temporally Consistent Video Super-Resolution with Video Diffusion Prior
Janghyeok Han (POSTECH), Sunghyun Cho (POSTECH)
Super ResolutionTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Propose a VSR method called DC-VSR based on video diffusion models, which can generate high-resolution videos that are coherent in both space and time and rich in details.
DeepMill: Neural Accessibility Learning for Subtractive Manufacturing
Fanchao Zhong (Shandong University), Haisen Zhao (Shandong University)
SegmentationConvolutional Neural NetworkGraph Neural NetworkSupervised Fine-TuningContrastive LearningPoint CloudMesh
🎯 What it does: Proposes DeepMill, a neural model based on octree convolutional networks, for rapidly predicting the reachability and severe occluded regions of cutting tools in subtractive manufacturing;
DeFillet: Detection and Removal of Fillet Regions in Polygonal CAD Models
Jingen Jiang, Wenping Wang
OptimizationMesh
🎯 What it does: A DeFillet algorithm is proposed to detect and remove fillet regions in polygonal CAD models.
Deformable Beta Splatting
Rong Liu (University of Southern California), A. Feng
OptimizationComputational EfficiencyRepresentation LearningNeural Radiance FieldGaussian SplattingImagePoint Cloud
🎯 What it does: Propose the Deformable Beta Splatting method, using deformable Beta kernels and spherical Beta encoding to achieve high-quality, real-time illumination field rendering.
Designing 3D Anisotropic Frame Fields with Odeco Tensors
Haikuan Zhu (Wayne State University), Zichun Zhong (Wayne State University)
OptimizationPoint CloudMeshOrdinary Differential Equation
🎯 What it does: This paper proposes a method that utilizes the odeco tensor representation to automatically optimize orientation and anisotropy ratios on three-dimensional volume grids, generating smooth and boundary-geometry-conforming high-plasticity tensor fields.
Designing Pin-pression Gripper and Learning its Dexterous Grasping with Online In-hand Adjustment
He Xiao, Kai Xu (National University Of Defense Technology)
Robotic IntelligenceReinforcement LearningPoint CloudMesh
🎯 What it does: Designed and implemented a Pin‑pression gripper capable of online adjustment of fingertip shape, and learned its dynamic strategy for grasping and in-hand repositioning through reinforcement learning.
DesignManager: An Agent-Powered Copilot for Designers to Integrate AI Design Tools into Creative Workflows
W. You, Lingyun Sun
AI Code AssistantReinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AIPrompt EngineeringImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Developed an AI-driven collaborative design support system called DesignManager, which helps designers collaborate with generative AI during the creative process, and provides node-based visualization, conversational interaction, and context management based on an agent framework.
Differentiable Geometric Acoustic Path Tracing using Time-Resolved Path Replay Backpropagation
U. Finnendahl, Marc Alexa
OptimizationDiffusion modelPhysics RelatedAudio
🎯 What it does: This paper proposes a differentiable geometric acoustics path tracing system, which uses time-resolved path replay backpropagation to simulate and compute derivatives of the sound energy spectrum, supporting gradient computation with respect to any parameters such as materials, sources, microphones, and scene geometry;
Diffuse-CLoC: Guided Diffusion for Physics-based Character Look-ahead Control
Xiaoyu Huang (University of California, Berkeley), Farbod Farshidian (RAI Institute)
GenerationOptimizationRobotic IntelligenceTransformerReinforcement LearningDiffusion modelPoint CloudMeshSequentialPhysics Related
🎯 What it does: Propose a physics-driven character control framework called Diffuse-CLoC, which utilizes a joint diffusion model to simultaneously generate future states and actions, achieving high-quality physically feasible motion generation without requiring a training set for re-tuning.
Diffusion as Shader: 3D-aware Video Diffusion for Versatile Video Generation Control
Zekai Gu, Yuan Liu
GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelAuto EncoderContrastive LearningVideoPoint CloudMesh
🎯 What it does: This paper proposes Diffusion as Shader (DaS), a 3D-aware video diffusion model that utilizes 3D tracked videos as control signals. It can achieve various video generation control tasks, such as animating meshes to videos, camera control, motion transfer, and object manipulation, within a single framework.
Digital Animation of Powder-Snow Avalanches
F. Nascimento, A. Paiva
Physics Related
🎯 What it does: A physics-based multi-layer model was constructed to visualize and simulate animation of powder snow avalanches on complex terrain using the finite volume method.
Discrete Torsion of Connection Forms on Simplicial Meshes
Theo Braune, M. Desbrun
MeshPhysics Related
🎯 What it does: A new discrete Levi-Civita connection is proposed, and based on this, discrete curvature tensors (discrete torsion) and torsion control for discrete connections are defined;
Divide-and-Conquer Embedding
Yuan-yuan Cheng, Xiao-Ming Fu
Mesh
🎯 What it does: Propose an exact method to embed a disk-topology triangulation into an arbitrary convex polygon, using a divide-and-conquer approach to decompose the problem into subnets mapped to sub-convex polygons, ultimately achieving a natural embedding of each triangle into the corresponding three-sided polygon.
Drag Your Gaussian: Effective Drag-Based Editing with Score Distillation for 3D Gaussian Splatting
Yansong Qu (Xiamen University), Rongrong Ji (Xiamen University)
GenerationData SynthesisOptimizationConvolutional Neural NetworkTransformerSupervised Fine-TuningPrompt EngineeringDiffusion modelScore-based ModelNeural Radiance FieldContrastive LearningGaussian SplattingImagePoint CloudMesh
🎯 What it does: This paper proposes DYG, a drag-based 3D Gaussian Splatting (3DGS) editing method, allowing users to perform fine-grained geometric editing on 3D scenes through 3D masks and control points;
DreamMask: Boosting Open-vocabulary Panoptic Segmentation with Synthetic Data
Yuanpeng Tu (HKU), Hengshuang Zhao (HKU)
SegmentationData SynthesisTransformerLarge Language ModelDiffusion modelContrastive LearningImageText
🎯 What it does: Propose DreamMask, a method that generates high-quality synthetic samples by leveraging LLM and diffusion models, and significantly improves the performance of open-vocabulary segmentation models on new categories through multi-stage filtering and synthetic-real alignment training.
Dress-1-to-3: Single Image to Simulation-Ready 3D Outfit with Diffusion Prior and Differentiable Physics
Xuan Li (University of California Los Angeles), Chenfanfu Jiang (University of Utah)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingOptical FlowImagePoint CloudMesh
🎯 What it does: Reconstruct a physically simulatable and disassemblable 3D garment from a single in-the-wild image, and automatically generate corresponding sewing patterns, materials, and animations.
Dual-Band Feature Fusion for Neural Global Illumination with Multi-Frequency Reflections
Shaohua Mo, Yuchi Huo
OptimizationComputational EfficiencyDiffusion modelNeural Radiance FieldImagePhysics Related
🎯 What it does: Proposes a neural global illumination method based on dual-frequency feature fusion, supporting multi-frequency reflection in dynamic scenes
DualMS: Implicit Dual-Channel Minimal Surface Optimization for Heat Exchanger Design
Weizheng Zhang (Shandong University), Qiang Du (Institute of Engineering Thermophysics, Chinese Academy of Sciences)
OptimizationGraph Neural NetworkDiffusion modelAuto EncoderContrastive LearningGaussian SplattingPoint CloudMeshGraphPhysics RelatedRetrieval-Augmented Generation
🎯 What it does: Propose the DualMS framework, which optimizes the separation surface through dual-channel minimal surface design for free-form heat exchangers, aiming to achieve a balance between maximizing heat transfer efficiency and minimizing pressure drop.
DuetGen: Music Driven Two-Person Dance Generation via Hierarchical Masked Modeling
Anindita Ghosh (DFKI), Chuan Guo (Snap Inc)
GenerationPose EstimationTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningVideoSequentialAudio
🎯 What it does: Designed the DuetGen framework, which can automatically generate synchronized and coordinated two-person dances based on music.
Dynamic Concepts Personalization from Single Videos
Rameen Abdal (Snap Research), Kfir Aberman (Snap Research)
GenerationData SynthesisTransformerSupervised Fine-TuningPrompt EngineeringDiffusion modelVideoText
🎯 What it does: Personalized learning of dynamic concepts from a single video to generate videos with high-fidelity appearance and motion.
Dynamic Mesh Processing on the GPU
Ahmed H. Mahmoud, John D. Owens
OptimizationComputational EfficiencyMesh
🎯 What it does: Proposes a fully GPU-based dynamic triangle mesh processing system, which utilizes mesh partitioning into small blocks and updates them in shared memory, employs speculative parallel processing to reduce rollbacks, and introduces a programming model with the cavity operator to uniformly handle dynamic updates.
Echoes of the Coliseum: Towards 3D Live streaming of Sports Events
Junkai Huang, Fernando de la Torre
GenerationData SynthesisPose EstimationComputational EfficiencyConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingOptical FlowVideoPoint CloudMeshBenchmark
🎯 What it does: Propose the LiveSplats framework to achieve real-time, realistic 3D reconstruction of live sports events, leveraging high-performance 3D Gaussian Splatting and interactive, freely explorable 3D experiences through distributed processing and load balancing.
EDGE: Epsilon-Difference Gradient Evolution for Buffer-Free Flow Maps
Zhiqi Li, Bo Zhu
OptimizationComputational EfficiencyOptical Flow
🎯 What it does: Propose the EDGE method, which utilizes Hermite interpolation for unbuffered flow graph computation on grids, and combines Gradient Evolution with a tetrahedral-based Epsilon Difference scheme to compute high-order derivatives, significantly reducing memory usage;
EditDuet: A Multi-Agent System for Video Non-Linear Editing
Marcelo Sandoval-Castañeda (TTI-Chicago), Fabian Caba Heilbron (Adobe)
GenerationData SynthesisTransformerLarge Language ModelAgentic AIVision Language ModelVideoTextRetrieval-Augmented Generation
🎯 What it does: Propose a multi-agent-based automatic video nonlinear editing system called EditDuet, which utilizes the collaboration between an editor and a critic LLM to iteratively generate B-roll video sequences;
Elastic Locomotion with Mixed Second-order Differentiation
Siyuan Shen (Zhejiang University), Yin Yang (University of Utah)
OptimizationRobotic IntelligenceMesh
🎯 What it does: A framework for elastic maneuvering based on a hybrid second-order derivative was constructed, utilizing inverse simulation and muscle activation optimization to achieve control of soft bodies toward high-level motion goals.
Elevating 3D Models: High-Quality Texture and Geometry Refinement from a Low-Quality Model
Nuri Ryu (POSTECH), Sunghyun Cho (POSTECH)
RestorationGenerationData SynthesisTransformerDiffusion modelScore-based ModelNeural Radiance FieldGaussian SplattingImagePoint CloudMeshStochastic Differential Equation
🎯 What it does: Propose the Elevate3D framework to achieve dual high-quality refinement of texture and geometry for low-quality 3D models
ELGAR: Expressive Cello Performance Motion Generation for Audio Rendition
Zhiping Qiu (Central Conservatory of Music), Qionghai Dai (Tsinghua University)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningVideoSequentialAudio
🎯 What it does: Generate complete cello performance actions directly from audio, including hand details and bow movements, forming a full-body interactive performance.
Escher Tile Deformation via Closed-Form Solution
Crane He Chen (Industrial Light & Magic), Vladimir G. Kim (Adobe Research)
GenerationData SynthesisPrompt EngineeringDiffusion modelAuto EncoderContrastive LearningOptical FlowImagePoint CloudMeshReview/Survey Paper
🎯 What it does: Propose a real-time interactive Escher texture deformation method that can simultaneously deform the boundaries and internal textures of tiles while preserving all 17 wallpaper group translational symmetries, avoiding gaps or overlaps.
EVA: Expressive Virtual Avatars from Multi-view Videos
Hendrik Junkawitsch (Max Planck Institute for Informatics), Marc Habermann (Max Planck Institute for Informatics)
GenerationData SynthesisPose EstimationConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkGaussian SplattingImageVideoPoint CloudMesh
🎯 What it does: Propose the EVA framework to generate photorealistic virtual avatars from multi-view videos, enabling full-body controllability, real-time performance, and independent expression of body, hands, and face.
FaceExpressions-70k: A Dataset of Perceived Expression Differences
Avinab Saha, A. Bovik
RecognitionSupervised Fine-TuningContrastive LearningImage
🎯 What it does: Propose and collect the FaceExpressions-70k dataset, which is oriented towards expression difference perception, containing 70,500 pairs of subjective comparisons of expressions, and demonstrate that it can be used to train perception difference models and guide decision-making on latency and sampling rates for facial avatars.
Facial Appearance Capture at Home with Patch-Level Reflectance Prior
Yuxuan Han (Tsinghua University), Feng Xu (Tsinghua University)
Image TranslationRestorationGenerationConvolutional Neural NetworkTransformerDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderImageVideoMesh
🎯 What it does: A home-based facial appearance capture method based on a Patch-level diffusion model was developed, which reconstructs high-quality illumination, geometry, and reflectance properties using a single smartphone plus flashlight video.
Facial Microscopic Structures Synthesis from a Single Unconstrained Image
Youyang Du, Beibei Wang
RestorationGenerationGraph Neural NetworkDiffusion modelImage
🎯 What it does: A differentiable graph neural network framework was developed for synthesizing facial microstructures (fine wrinkles and pores) from a single unconstrained blurry facial image, and it can be combined with existing macro facial detail reconstruction methods.
Faraday Cage Estimation of Normals for Point Clouds and Ribbon Sketches
Daniel Scrivener, Faraday
OptimizationComputational EfficiencyRepresentation LearningScore-based ModelPoint CloudMeshPhysics Related
🎯 What it does: Propose the FaCE method, which utilizes the Faraday cage effect to estimate normals from unoriented point clouds and VR ribbon sketches
FashionComposer: Compositional Fashion Image Generation
Sihui Ji (University of Hong Kong), Hengshuang Zhao (University of Hong Kong)
Image TranslationImage HarmonizationGenerationData SynthesisPose EstimationTransformerLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Proposes FashionComposer, a multi-modal compositional fashion image generation framework based on diffusion models, which supports various inputs such as text, SMPL joints, facial features, and clothing, and can synthesize multiple pieces of clothing at once, achieving multi-task capabilities such as virtual try-on and photo album generation.
Fast But Accurate: A Real-Time Hyperelastic Simulator with Robust Frictional Contact
Ziqiu Zeng (University of Strasbourg), Zhongkai Zhang
OptimizationComputational EfficiencyRobotic IntelligenceMesh
🎯 What it does: Developed a GPU-friendly real-time implicit elastic material simulation framework that can simultaneously handle large deformations, non-penetration constraints, and frictional contact.