arXivSub Start free trial

SIGGRAPH Asia 2023 Papers — Page 2

ACM SIGGRAPH Asia (Transactions on Graphics) · 224 papers

High-Fidelity and Real-Time Novel View Synthesis for Dynamic Scenes

Haotong Lin (Zhejiang University), Xiaowei Zhou (Zhejiang University)

GenerationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkTransformerDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingImageVideoBenchmark

🎯 What it does: Propose a hybrid scene representation called Im4D, which integrates grid-based four-dimensional geometry with multi-view image rendering for high-fidelity real-time view synthesis in dynamic scenes.

High-Order Moment-Encoded Kinetic Simulation of Turbulent Flows

Wei Li, Mathieu Desbrun

Diffusion modelScore-based ModelPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Propose a high-order moment encoding Lagrangian Boltzmann method solver (HOME-LBM) for incompressible fluid simulation.

Holographic Near-eye Display with Real-time Embedded Rendering

Antonin Gilles, P. Gioia

Neural Radiance FieldOptical FlowImage

🎯 What it does: A wearable augmented reality headset with binocular vision support and real-time embedded full-color holographic rendering was proposed.

HyperDreamer: Hyper-Realistic 3D Content Generation and Editing from a Single Image

Tong Wu (Chinese University of Hong Kong), Ziwei Liu (Nanyang Technological University)

GenerationData SynthesisConvolutional Neural NetworkTransformerPrompt EngineeringDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningImagePoint CloudMesh

🎯 What it does: Based on a single image, HyperDreamer is proposed to generate panoramic, renderable, and editable high-resolution 3D models.

IconShop: Text-Guided Vector Icon Synthesis with Autoregressive Transformers

Rong Wu (City University of Hong Kong), Jing Liao (City University of Hong Kong)

GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageTextSequential

🎯 What it does: Developed IconShop, a text-guided vector icon (SVG) synthesis method based on autoregressive Transformer, capable of generating black-and-white monochrome icons according to keywords or natural language descriptions, and supporting multiple applications such as icon editing, interpolation, and semantic composition.

Inovis: Instant Novel-View Synthesis

Mathias Harrer, Tim Weyrich

GenerationData SynthesisDiffusion modelNeural Radiance FieldAuto EncoderGaussian SplattingImage

🎯 What it does: Developed a neural rendering system called Inovis, which utilizes a space-time upsampling architecture to recompose and merge multi-view camera images, and synthesizes new views during inference.

Input-Dependent Uncorrelated Weighting for Monte Carlo Denoising

Jonghee Back, Bochang Moon

RestorationConvolutional Neural NetworkDiffusion modelScore-based ModelImage

🎯 What it does: Propose a new Monte Carlo denoising framework based on input-uncorrelated weights.

Interaction-Driven Active 3D Reconstruction with Object Interiors

Zihao Yan (Shenzhen University), Hui Huang (Shenzhen University)

Robotic IntelligenceTransformerNeural Radiance FieldAuto EncoderContrastive LearningSimultaneous Localization and MappingOptical FlowPoint CloudMeshGraph

🎯 What it does: Propose a 3D reconstruction method based on robot interaction and active scanning, which can simultaneously acquire the external and internal geometry of objects and identify movable parts and their motion parameters.

Interactive Story Visualization with Multiple Characters

Yuan Gong (Tsinghua University), Yujiu Yang (Tsinghua University)

GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageVideoText

🎯 What it does: A complete interactive story visualization system is proposed, which can generate multi-character images based on textual stories and produce animations.

Intrinsic Harmonization for Illumination-Aware Image Compositing

Chris Careaga (Simon Fraser University), Yağız Aksoy (Simon Fraser University)

Image HarmonizationRestorationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningImage

🎯 What it does: Propose a self-supervised, illumination-aware image fusion method based on intrinsic images (reflectance and shadows), which can simultaneously correct color and illumination mismatches between foreground and background in wild scenes.

iNVS: Repurposing Diffusion Inpainters for Novel View Synthesis

Yash Kant (University of Toronto), Igor Gilitschenski (University of Toronto)

GenerationData SynthesisDepth EstimationTransformerSupervised Fine-TuningDiffusion modelGaussian SplattingImagePoint CloudMesh

🎯 What it does: Propose the iNVS method, which uses a pre-trained 2D inpainting diffusion model to synthesize new views from a single input image.

Joint Sampling and Optimisation for Inverse Rendering

Martin Bálint (Max Planck Institute for Informatics), G. Singh

OptimizationComputational EfficiencyMeta LearningDiffusion modelScore-based ModelNeural Radiance FieldContrastive LearningGaussian SplattingImageMesh

🎯 What it does: Propose a joint sampling and optimization meta-estimation framework for efficient gradient updates in noisy inverse rendering.

K-Surfaces: Bézier-Splines Interpolating at Gaussian Curvature Extrema

T. Djuren, Marc Alexa

OptimizationMesh

🎯 What it does: Propose the K-surfaces technique, which achieves interpolation of Bezier-spline surface control points at extreme points of Gaussian curvature by learning the inverse mapping, and ensures cross-patch continuity in interactive modeling through iterative optimization.

Kirchhoff-Love Shells with Arbitrary Hyperelastic Materials

Jiahao Wen, J. Barbič

Physics Related

🎯 What it does: Derived the mechanical energy of arbitrary three-dimensional hyperelastic materials in Kirchhoff-Love thin shells, explicitly separating in-plane stretching and bending terms, avoiding numerical integration.

Learning Based 2D Irregular Shape Packing

Zeshi Yang (LightSpeed Studios), Xifeng Gao (LightSpeed Studios)

OptimizationConvolutional Neural NetworkGraph Neural NetworkReinforcement LearningDiffusion modelAuto EncoderContrastive LearningMesh

🎯 What it does: Proposes a learning-based 2D irregular shape packing method for efficiently packing UV patches into a texture atlas;

Learning Contact Deformations with General Collider Descriptors

Cristian Romero, M. Otaduy

Computational EfficiencyRepresentation LearningRobotic IntelligenceGenerative Adversarial NetworkContrastive LearningMesh

🎯 What it does: Proposed a learning-based generic collider descriptor to achieve rich contact deformation simulations on simplified deformable models.

Learning Gradient Fields for Scalable and Generalizable Irregular Packing

Tianyang Xue (Shandong University), Baoquan Chen (Peking University)

OptimizationGraph Neural NetworkTransformerDiffusion modelScore-based ModelPoint CloudMeshStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Proposes a gradient field learning method based on score-based diffusion models for efficient, scalable, and generalizable 2D irregular packing problems.

Learning the Geodesic Embedding with Graph Neural Networks

Bo Pang (Peking University), Peng-Shuai Wang (Peking University)

Computational EfficiencyRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningMesh

🎯 What it does: Propose a GeGnn method based on graph neural networks to learn geodesic distance embeddings between two points on a surface, enabling constant-time distance queries after a single forward pass.

LiCROM: Linear-Subspace Continuous Reduced Order Modeling with Neural Fields

Yue Chang, E. Grinspun (University of Toronto)

OptimizationComputational EfficiencyRepresentation LearningGraph Neural NetworkDiffusion modelScore-based ModelRectified FlowNeural Radiance FieldAuto EncoderContrastive LearningPoint CloudMeshGraphPhysics Related

🎯 What it does: A discretization-agnostic linear subspace continuous ROM (LiCROM) was developed, which uses neural implicit fields to represent continuous displacement bases. After training, it can perform real-time simulations on different grids, topologies, and unseen geometries.

Light-Efficient Holographic Illumination for Continuous-Wave Time-of-Flight Imaging

D. Chan, Matthew O'Toole

OptimizationComputational EfficiencyOptical FlowImagePhysics Related

🎯 What it does: Propose a method that utilizes a holographic light source to redistribute illumination in order to improve the dynamic range of continuous wave TOF phase cameras

LitNeRF: Intrinsic Radiance Decomposition for High-Quality View Synthesis and Relighting of Faces

Kripasindhu Sarkar, Abhimitra Meka

GenerationData SynthesisDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingOptical FlowImageMesh

🎯 What it does: Proposed a sparse capture device utilizing 15 cameras and 15 lights, combining a hybrid method of neural voxel representation with traditional multi-view stereo reconstruction for high-quality facial 3D view synthesis and relighting.

LiveNVS: Neural View Synthesis on Live RGB-D Streams

Laura Fink (Friedrich-Alexander-Universität Erlangen-Nürnberg), Marc Stamminger (Friedrich-Alexander-Universität Erlangen-Nürnberg)

GenerationPose EstimationDepth EstimationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingOptical FlowImageVideoPoint Cloud

🎯 What it does: Achieve real-time, unprocessed 3D scene visualization through neural viewpoint synthesis using real-time RGB-D streams.

Locally-Adaptive Level-of-Detail for Hardware-Accelerated Ray Tracing

J. Haydel, Larry Seiler

Computational EfficiencyNeural Radiance FieldOptical FlowMesh

🎯 What it does: Proposes a local adaptive detail (LOD) technique for hardware-accelerated ray tracing, which can reduce memory bandwidth consumption during the ray traversal process, thereby decreasing energy consumption and rendering time.

Lock-free Vertex Clustering for Multicore Mesh Reduction

Nima Fathollahi, S. Chester

OptimizationComputational EfficiencyPoint CloudMesh

🎯 What it does: Proposed a lock-free parallel algorithm called P-Weld for vertex clustering to achieve mesh simplification.

Low-Light Image Enhancement with Wavelet-Based Diffusion Models

Hailin Jiang, Shuaicheng Liu

RestorationSuper ResolutionConvolutional Neural NetworkTransformerDiffusion modelImage

🎯 What it does: Proposed a condition diffusion model called DiffLL based on wavelet transform for low-light image enhancement.

Manifold Path Guiding for Importance Sampling Specular Chains

Zhimin Fan (Nanjing University), Ling-Qi Yan (University of California, Santa Barbara)

OptimizationComputational EfficiencyDiffusion modelScore-based ModelFlow-based ModelImagePhysics RelatedRetrieval-Augmented Generation

🎯 What it does: This paper proposes an unbiased Monte Carlo light path sampling method based on manifold path guidance, specifically designed for importance sampling of specular chains of arbitrary length.

MatFusion: A Generative Diffusion Model for SVBRDF Capture

Sam Sartor (William & Mary), Pieter Peers (William & Mary)

GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelScore-based ModelAuto EncoderImage

🎯 What it does: Proposed a generative diffusion model called MatFusion for estimating spatially varying BRDF (SVBRDF) from photographs, with conditional improvements under three lighting conditions.

MCNeRF: Monte Carlo Rendering and Denoising for Real-Time NeRFs

Kunal Gupta, Sai Bi

Computational EfficiencyNeural Radiance Field

🎯 What it does: Propose MCNeRF, a rendering algorithm based on Monte Carlo, which can accelerate any NeRF representation, and estimates pixel radiance with a small number of MLP evaluations, followed by denoising using an image-space denoiser.

Meshes with Spherical Faces

Martin Kilian, H. Pottmann

Diffusion modelScore-based ModelOptical FlowMeshStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Systematically describes meshes with spherical faces, achieving the action of the Mobius transformation group through circular arc edges, providing a method for clustering spherical panels by radius, generating spherical meshes with geometric support structures, characterizing non-Euclidean geometric features for triangular and hexagonal combinations respectively, and generating visually convex hexagons through tangent sphere intersections; additionally, extending all combinations of circular face meshes into spherical meshes, filling spherical caps, and providing a remeshing scheme for quadrilateral spherical meshes, further improving overall surface perception by minimizing the dihedral angles between adjacent spheres.

MetaLayer: A Meta-Learned BSDF Model for Layered Materials

Jie Guo, Ling-Qi Yan

Meta LearningNeural Radiance FieldAuto Encoder

🎯 What it does: Proposes MetaLayer, a new method for modeling and rendering layered materials using meta-learning;

Metric Optimization in Penner Coordinates

Ryan Capouellez (New York University), D. Zorin

OptimizationMesh

🎯 What it does: A method is proposed that utilizes Penner coordinates to perform metric optimization and interpolation across the entire cone metric space, capable of minimizing various distortion energies while satisfying angular constraints, enabling a smooth transition from full isometric mapping to isometric optimization mapping, and generating high-quality parameterizations.

MIPS-Fusion: Multi-Implicit-Submaps for Scalable and Robust Online Neural RGB-D Reconstruction

Yijie Tang (National University of Defense Technology), Kaiyang Xu

Pose EstimationDepth EstimationAutonomous DrivingOptimizationComputational EfficiencyDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningGaussian SplattingSimultaneous Localization and MappingOptical FlowImageVideoPoint Cloud

🎯 What it does: Propose MIPS-Fusion, an online RGB-D reconstruction framework based on multiple implicit subgraphs.

MOCHA: Real-Time Motion Characterization via Context Matching

Deok-Kyeong Jang (KAIST), Sung-Hee Lee (KAIST)

GenerationRepresentation LearningGraph Neural NetworkTransformerAuto EncoderContrastive LearningVideoSequential

🎯 What it does: Proposed a real-time motion representation framework called MOCHA, which can convert unstyled motions into motion styles and body types of specified characters.

Multi-color Holograms Improve Brightness in Holographic Displays

Koray Kavaklı (Koç University), K. Akşit

OptimizationDiffusion modelNeural Radiance FieldOptical FlowImageVideoPhysics Related

🎯 What it does: A multi-color holographic driving scheme was designed, utilizing multi-wavelength lasers to simultaneously illuminate a single-phase grating, and jointly optimizing the phase map and the laser intensity of each sub-frame through gradient descent to achieve brighter holographic displays.

Multiple-bounce Smith Microfacet BRDFs using the Invariance Principle

Yuang Cui (Anhui Science and Technology University), Beibei Wang (Nankai University)

Diffusion modelImageBenchmarkPhysics Related

🎯 What it does: Propose a multi-bounce Smith microsurface BRDF model derived using the principle of invariance, and provide a concise paragraph function.

Multisource Holography

Grace Kuo (Reality Labs Research, Meta), N. Matsuda

GenerationOptimizationDiffusion modelOptical FlowImagePhysics Related

🎯 What it does: A multi-source holographic display architecture is proposed, which generates single-frame noise-free high-resolution holograms using multiple light sources and two spatial light modulators.

MuscleVAE: Model-Based Controllers of Muscle-Actuated Characters

Yusen Feng (Peking University), Libin Liu (Peking University)

Robotic IntelligenceReinforcement LearningAuto EncoderWorld ModelVideoSequential

🎯 What it does: Proposes a muscle-driven character simulation and control framework based on the Hill muscle model and the 3CC-r fatigue model, and designs a muscle space PD control and a VAE-based generator controller called MuscleVAE.

MyStyle++: A Controllable Personalized Generative Prior

Libing Zeng (Texas A&M University), N. Kalantari

GenerationData SynthesisSupervised Fine-TuningDiffusion modelGenerative Adversarial NetworkContrastive LearningImage

🎯 What it does: A controllable personalized generation prior (MyStyle++) was constructed to achieve precise control over facial attributes while preserving the identity characteristics of the target person.

Neural Caches for Monte Carlo Partial Differential Equation Solvers

Zilu Li, Steve Marschner

OptimizationComputational EfficiencyNeural Radiance FieldPhysics RelatedStochastic Differential Equation

🎯 What it does: Combine neural network caching mechanisms with Monte Carlo PDE solvers (such as Walk-on-Spheres), by training neural fields to approximate PDE solutions and using them as a cache to reduce variance.

Neural Categorical Priors for Physics-Based Character Control

Qing Zhu, Lei Han (Tencent Robotics X)

GenerationData SynthesisRobotic IntelligenceTransformerReinforcement LearningPrompt EngineeringDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningVideoSequentialPhysics Related

🎯 What it does: Developed a generative model trained using Neural Categorical Priors (NCP) for physical simulation character control, and enhanced motion diversity through prior shifting.

Neural Collision Fields for Triangle Primitives

Ryan S. Zesch, D. Levin

Neural Radiance FieldOptical FlowMeshPhysics Related

🎯 What it does: Propose a neural collision field to replace contact point sampling in physical simulation, and handle collisions between triangular meshes.

Neural Field Convolutions by Repeated Differentiation

Ntumba Elie Nsampi (MPI Informatik), Thomas Leimkühler (MPI Informatik)

Computational EfficiencyRepresentation LearningConvolutional Neural NetworkDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningImageVideoPoint CloudMeshTime SeriesStochastic Differential EquationOrdinary Differential EquationAudio

🎯 What it does: Propose a method that efficiently convolves continuous neural fields with piecewise polynomial kernels by utilizing repeated differentiation and integration; by repeatedly differentiating the kernel, sparse Dirac point sets are obtained, and a 'repeated integration field' is trained to evaluate the kernel convolution results.

Neural Gradient Learning and Optimization for Oriented Point Normal Estimation

Qing Li (Tsinghua University), Zhizhong Han (Wayne State University)

Pose EstimationDepth EstimationOptimizationRepresentation LearningGraph Neural NetworkContrastive LearningPoint Cloud

🎯 What it does: Propose a two-stage framework based on deep learning—Neural Gradient Learning (NGL) and Gradient Vector Optimization (GVO)—for directly estimating consistently oriented normals from noisy, dense, and diverse point clouds.

Neural Metamaterial Networks for Nonlinear Material Design

Yue Li (ETH Zürich), Bernhard Thomaszewski (ETH Zürich)

OptimizationComputational EfficiencyGraph Neural NetworkSpiking Neural NetworkTransformerDiffusion modelAuto EncoderContrastive LearningMeshGraphTabularTime SeriesPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Proposed a neuro-morphological material network (NMN) for efficient inverse design of nonlinear morphological materials, directly optimizing structural parameters to match target stress-strain curves, directional stiffness, or Poisson's ratio curves.

Neural Motion Graph

Hongyu Tao, Weiwei Xu

GenerationPose EstimationGraph Neural NetworkNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkGraph

🎯 What it does: Propose a neural motion graph, using separate neural nodes for each action type and a transition network with a lightweight control module, achieving scalable and highly controllable human motion synthesis.

Neural Packing: from Visual Sensing to Reinforcement Learning

Juzhan Xu (Shenzhen University), Ruizhen Hu (Shenzhen University)

Robotic IntelligenceConvolutional Neural NetworkTransformerSupervised Fine-TuningReinforcement LearningImagePoint CloudBenchmark

🎯 What it does: Propose a complete transportation and packaging (TAP) learning framework that uses RGBD visual perception to acquire source box information and target container height maps. A neural network is used to simultaneously determine which boxes to move and their placement positions, and motion planning is employed to realize real robotic arm execution, ultimately achieving compact and efficient three-dimensional packing.

Neural Point-based Volumetric Avatar: Surface-guided Neural Points for Efficient and Photorealistic Volumetric Head Avatar

Cong Wang (Tsinghua University), Songiie Zhang

GenerationDepth EstimationComputational EfficiencyRepresentation LearningDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningGaussian SplattingImageVideoPoint CloudMesh

🎯 What it does: Propose a voxelized facial avatar representation based on movable neural points (NPVA), achieving high-fidelity and animatable facial rendering by dynamically distributing neural points on the UV surface of the target expression.

Neural Spectro-polarimetric Fields

Youngchan Kim (POSTECH), Seung-Hwan Baek (POSTECH)

RestorationData SynthesisCompressionDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningOptical FlowImageMultimodalityPhysics RelatedStochastic Differential Equation

🎯 What it does: This paper proposes a neural network model called NeSpoF for scene reconstruction and rendering with high-dimensional optical parameters (wavelength, polarization, spatial, and directional), and achieves the acquisition and reconstruction of multi-view hyperspectral polarized images.

Neural Stochastic Poisson Surface Reconstruction

Silvia Sellán, Alec Jacobson (University of Toronto)

OptimizationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkRecurrent Neural NetworkDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderContrastive LearningGaussian SplattingPoint CloudMesh

🎯 What it does: Propose a random masking Poisson reconstruction framework based on neural networks, representing both mean and covariance with multi-layer perceptrons, supporting efficient iterative updates, next best view planning, and seamless integration into the scanning process.

Neural Stress Fields for Reduced-order Elastoplasticity and Fracture

Zeshun Zong (UCLA), Peter Yichen Chen (MIT)

OptimizationComputational EfficiencyDiffusion modelScore-based ModelFlow-based ModelRectified FlowNeural Radiance FieldAuto EncoderPoint CloudMeshPhysics Related

🎯 What it does: Proposed a low-rank modeling framework that combines neural networks with physical models for efficient simulation of elasto-plasticity and cracking.

Neural-Singular-Hessian: Implicit Neural Representation of Unoriented Point Clouds by Enforcing Singular Hessian

Zixiong Wang (Shandong University), Changhe Tu (Shandong University)

RestorationRepresentation LearningDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingPoint Cloud

🎯 What it does: This paper proposes a new implicit neural surface reconstruction method that utilizes the singular Hessian constraint (making the Hessian determinant zero for points near the surface) to enhance the reconstruction of unoriented point clouds;

NodeGit: Diffing and Merging Node Graphs

Eduardo Rinaldi, Fabio Pellacini

OptimizationAI Code AssistantGraph Neural NetworkDiffusion modelScore-based ModelGraph

🎯 What it does: Proposed a practical algorithm for computing differences (diff) and merging (merge) of procedural node graphs, achieving more accurate difference information by directly versioning the graph.

Non-Newtonian ViRheometry via Similarity Analysis

Mitsuki Hamamichi, Yonghao Yue

OptimizationVideoPhysics Related

🎯 What it does: By using unknown materials in dam-break or column collapse experiments and capturing videos, then utilizing simulation optimization to estimate three parameters (σY, n, η), thus estimating the Herschel-Bulkley parameters.

Nonlinear Ray Tracing for Displacement and Shell Mapping

Shinji Ogaki

Neural Radiance FieldMesh

🎯 What it does: A novel and efficient method is proposed, which fully represents nonlinear rays using quadratic rational functions in the texture space, performs accelerated structure traversal and intersection testing on micro-triangles, thereby simplifying the implementation of displacement mapping and smooth shell mapping without subdivision, and can handle degenerate base triangles in the UV space.

Object Motion Guided Human Motion Synthesis

Jiaman Li (Stanford University), C. K. Liu (Stanford University)

GenerationData SynthesisPose EstimationTransformerDiffusion modelScore-based ModelVideoPoint CloudMesh

🎯 What it does: Generating complete human motions from object interactions.

Online Scene CAD Recomposition via Autonomous Scanning

Changhao Li, Ligang Liu

Autonomous DrivingOptimizationRobotic IntelligenceSimultaneous Localization and MappingOptical FlowPoint CloudMeshRetrieval-Augmented Generation

🎯 What it does: This paper proposes an online scene CAD reconstruction method, which utilizes autonomous scanning and guides scene reconstruction through automatically optimized next best view (NBV) during a single scanning process, directly combining CAD models retrieved from a given dataset into results that match the actual scene geometry and layout.

OpenSVBRDF: A Database of Measured Spatially-Varying Reflectance

Xiaohe Ma, Hongzhi Wu

Neural Radiance FieldAuto EncoderContrastive LearningOptical FlowImageMeshBenchmark

🎯 What it does: Constructed and released the first large-scale measured spatially varying anisotropic reflectance database, containing 1000 high-quality near-planar SVBRDF samples, and provided high-resolution texture maps;

Optimal Design of Robotic Character Kinematics

Guirec Maloisel, Moritz Bächer

OptimizationRobotic Intelligence

🎯 What it does: Propose a technique that uses dynamic programming to simultaneously optimize the mechanical design and control parameters of a robot, and removes common motion loop constraints, enabling the editing of existing robot designs or the optimization of new designs to achieve target movements.

Perceptual error optimization for Monte Carlo animation rendering

Misa Korac, Gurprit Singh (Max Planck Institute for Informatics)

OptimizationConvolutional Neural NetworkDiffusion modelScore-based ModelContrastive LearningOptical FlowImageVideo

🎯 What it does: An optimized framework is designed for Monte Carlo animation rendering, achieving error distributions that exhibit blue noise characteristics in both spatial and temporal domains, thereby significantly improving perceptual quality.

Perceptual Requirements for World-Locked Rendering in AR and VR

Phillip Guan (Meta), Douglas Lanman (Meta)

Vision Language ModelSimultaneous Localization and MappingOptical FlowImageVideoMultimodalityReview/Survey PaperBenchmarkPhysics Related

🎯 What it does: This paper designs an AR/VR system that can achieve worldwide locked rendering (WLR), and evaluates the perceptual threshold of active audience to rendering camera position errors through psychophysical experiments.

Perceptually Adaptive Real-Time Tone Mapping

Taimoor Tariq, Alexandre Chapiro

Computational EfficiencyContrastive LearningImageVideo

🎯 What it does: Proposed a real-time perceptual contrast matching framework for mapping scene content to the dynamic range of the target display

Pose and Skeleton-aware Neural IK for Pose and Motion Editing

Dhruv Agrawal, Robert W. Sumner

Pose EstimationGraph Neural NetworkNeural Radiance FieldMeshGraph

🎯 What it does: Proposes a neural inverse kinematics method that can automatically complete the full 3D character pose through a small number of control points.

Power Plastics: A Hybrid Lagrangian/Eulerian Solver for Mesoscale Inelastic Flows

Ziyin Qu, F. Goes

OptimizationOptical FlowPhysics Related

🎯 What it does: A hybrid Lagrangian/Eulerian method was developed for simulating incompressible flows at the mesoscopic scale, capable of generating high-quality and volume-adaptive particle distributions.

Progressive Shell Qasistatics for Unstructured Meshes

J. Zhang, Danny M. Kaufman

OptimizationComputational EfficiencyMesh

🎯 What it does: Propose the Progressive Shell Quasistatics framework, extending progressive simulation to all input shell and plate geometries, supporting unstructured triangular meshes. Construct a fine-to-coarse hierarchy, design a nonlinear extension operator and shape-preserving upsampling method, ensuring no intersection and strain constraints, and efficiently reconstruct fine geometry on coarse grids.

Projective Sampling for Differentiable Rendering of Geometry

Ziyi Zhang, Wenzel Jakob

OptimizationComputational EfficiencyDiffusion modelNeural Radiance FieldMesh

🎯 What it does: Address the gradient bias caused by the discontinuity of boundary visibility in differentiable rendering, propose a boundary projection sampling strategy, and provide the corresponding theoretical analysis and algorithm implementation.

ProSpect: Prompt Spectrum for Attribute-Aware Personalization of Diffusion Models

Yu-xin Zhang, Changsheng Xu

GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelImageText

🎯 What it does: Proposes the ProSpect method, which achieves attribute-aware personalized image generation and editing by mapping a single reference image into multi-stage text embeddings, enabling the decoupling and recombination of visual attributes such as material, style, and layout.

PSDR-Room: Single Photo to Scene using Differentiable Rendering

Kai Yan, Shuang Zhao

GenerationData SynthesisOptimizationTransformerVision Language ModelDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingImageTextMeshRetrieval-Augmented Generation

🎯 What it does: Propose an end-to-end system called PSDR‑Room, which can automatically generate editable 3D scenes, including geometry, materials, and lighting, from a single indoor photo;

Quantum Ray Marching: Reformulating Light Transport for Quantum Computers

Logan Mosier, T. Hachisuka

Physics Related

🎯 What it does: Proposes a quantum ray marching method based on quantum random walks, achieving a complete quantum rendering pipeline capable of simulating light transport.

Reach For the Spheres: Tangency-aware surface reconstruction of SDFs

Silvia Sell'an, Oded Stein (University of Southern California)

OptimizationDiffusion modelScore-based ModelOptical FlowPoint CloudMeshStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: This paper proposes a global energy function that treats each SDF sample as a sphere and introduces tangent point constraints, achieving reconstruction from SDF to explicit triangle mesh via geometric flow gradient descent;

Real-time Height-field Simulation of Sand and Water Mixtures

Haozhe Su, Kui Wu

Optical FlowPhysics Related

🎯 What it does: Proposes a real-time simulation method for sand-water mixtures based on a height field.

Reconstructing Close Human Interactions from Multiple Views

Qing Shuai (Zhejiang University), Xiaowei Zhou (Zhejiang University)

Pose EstimationConvolutional Neural NetworkGraph Neural NetworkDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningOptical FlowImagePoint CloudBenchmark

🎯 What it does: Propose a multi-view learning framework that utilizes a 3D conditional convolutional network to reconstruct the 3D poses of closely interacting crowds from 2D heatmaps captured by multiple cameras, effectively handling occlusions and identity confusion without relying on real image-3D paired data.

Reconstruction of Machine-Made Shapes from Bitmap Sketches

Ivan Puhachov, Mikhail Bessmeltsev

RestorationGenerationOptimizationDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderImageMesh

🎯 What it does: Reconstructing 3D machine-manufactured shapes by separating patches from bitmap sketches and jointly optimizing geometry.

Rectifying Strip Patterns

Bolun Wang, H. Pottmann

OptimizationRectified Flow

🎯 What it does: Provides a shape design computation tool for flexible strips (rectifying strips) made from inextensible flat materials, capable of generating various patterns while satisfying constraints for specific applications such as grid structures or shading systems.

ReparamCAD: Zero-shot CAD Re-Parameterization for Interactive Manipulation

Milin Kodnongbua, Adriana Schulz

OptimizationComputational EfficiencyData-Centric LearningTransformerLarge Language ModelVision Language ModelTextMesh

🎯 What it does: Developed ReparamCAD zero-shot pipeline, which infers the meaningful variation space of shapes using pre-trained large language models and image models, and reparameterizes CAD programs based on this, enabling easy exploration along meaningful design axes.

Rerender A Video: Zero-Shot Text-Guided Video-to-Video Translation

Shuai Yang (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)

Image TranslationGenerationData SynthesisTransformerPrompt EngineeringDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningOptical FlowImageVideoText

🎯 What it does: This paper proposes a zero-training text-guided video-to-video translation framework, combining keyframe rendering with non-keyframe interpolation to achieve high-quality, temporally consistent video translation.

ReShader: View-Dependent Highlights for Single Image View-Synthesis

Avinash Paliwal (Texas A&M University), N. Kalantari (Texas A&M University)

GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningImagePoint Cloud

🎯 What it does: Propose to decompose single-image novel view synthesis into two steps: pixel re-coloring and pixel re-localization. After training a specialized re-coloring network, combine it with existing re-localization methods to generate new views with perspective-dependent highlights that move naturally.

RMIP: Displacement ray tracing via inversion and oblong bounding

T. Thonat, T. Boubekeur

Computational EfficiencyDiffusion modelOptical FlowMesh

🎯 What it does: Proposes an efficient displacement mapping inverse ray tracing method that directly finds ray-surface intersections in the displacement map space without pre-subdividing the triangular mesh.

Robust Zero Level-Set Extraction from Unsigned Distance Fields Based on Double Covering

Fei Hou (Chinese Academy of Sciences), Ying He (Nanyang Technological University)

OptimizationComputational EfficiencyRepresentation LearningMixture of ExpertsDiffusion modelAuto EncoderContrastive LearningOptical FlowPoint CloudMeshGraphStochastic Differential Equation

🎯 What it does: Propose a method called DoubleCoverUDF, which utilizes the double covering structure of r-offset and coverage mapping learning to robustly extract zero-level sets from the learned unsigned distance field (UDF), supporting both open and closed 3D surfaces.

RT-Octree: Accelerate PlenOctree Rendering with Batched Regular Tracking and Neural Denoising for Real-time Neural Radiance Fields

Zixi Shu, Lizhuang Ma

Computational EfficiencyDiffusion modelNeural Radiance FieldAuto EncoderGaussian SplattingImage

🎯 What it does: Propose RT-Octree and achieve faster real-time rendering through batch regularization tracking and neural denoising.

SAILOR: Synergizing Radiance and Occupancy Fields for Live Human Performance Capture

Zheng Dong, Rynson W. H. Lau

GenerationPose EstimationDepth EstimationConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderGaussian SplattingOptical FlowImageVideoPoint Cloud

🎯 What it does: Proposes the SAILOR method for generating high-quality human free-viewpoint videos from sparse RGBD live streams.

SAME: Skeleton-Agnostic Motion Embedding for Character Animation

Sunmin Lee, Jungdam Won

GenerationPose EstimationRepresentation LearningGraph Neural NetworkAuto EncoderContrastive LearningGraph

🎯 What it does: Propose a skeleton-agnostic motion embedding framework called SAME for solving various animation tasks

ScaNeRF: Scalable Bundle-Adjusting Neural Radiance Fields for Large-Scale Scene Rendering

Xiuchao Wu, Weiwei Xu

Pose EstimationDepth EstimationOptimizationComputational EfficiencyNeural Radiance FieldGaussian SplattingOptical FlowImagePoint Cloud

🎯 What it does: Proposes ScaNeRF, a large-scale scene rendering method that enhances the scalability and camera pose accuracy of beam-aligned neural radiance fields by using tile-based hybrid neural fields and parallel distributed optimization.

Scene-Aware Activity Program Generation with Language Guidance

Zejia Su, Ruizhen Hu

GenerationTransformerLarge Language ModelVision-Language-Action ModelWorld ModelImageVideoText

🎯 What it does: Proposes a scene-aware activity program generation method that combines language models with dynamic scene perception.

SeamlessNeRF: Stitching Part NeRFs with Gradient Propagation

Bingchen Gong (Chinese University of Hong Kong), Qi Dou (Chinese University of Hong Kong)

GenerationData SynthesisTransformerDiffusion modelScore-based ModelNeural Radiance FieldImagePoint Cloud

🎯 What it does: This paper proposes SeamlessNeRF, a gradient propagation-based NeRF seamless stitching method, which can smoothly fuse the appearance of multiple partial NeRFs through boundary alignment and gradient consistency;

Second-Order Finite Elements for Deformable Surfaces

Qiqin Le, Tao Du

OptimizationComputational EfficiencyMesh

🎯 What it does: A deformable surface simulation framework based on second-order triangular finite elements is proposed, and discretization methods for stretching, shearing, bending energy, and the mean curvature of surface triangular meshes are developed, integrating virtual node finite elements to achieve bidirectional coupling of shear rods.

Self-Calibrating, Fully Differentiable NLOS Inverse Rendering

Kiseok Choi (KAIST), Min H. Kim (KAIST)

OptimizationDiffusion modelScore-based ModelNeural Radiance FieldOptical FlowImagePoint CloudPhysics Related

🎯 What it does: This paper proposes an end-to-end differentiable NLOS inverse rendering pipeline, capable of self-calibrating imaging parameters and reconstructing hidden scene surface points, normals, and albedo;

SFLSH: Shape-Dependent Soft-Flesh Avatars

Pablo Ramón, M. Otaduy

GenerationData SynthesisDiffusion modelAuto EncoderContrastive LearningMesh

🎯 What it does: Proposed a multi-person soft tissue avatar model that maps body shape descriptors to geometric and mechanical parameters across the entire body, achieving shape-dependent soft avatars;

ShaDDR: Interactive Example-Based Geometry and Texture Generation via 3D Shape Detailization and Differentiable Rendering

Qimin Chen (Simon Fraser University), Hao Zhang (Simon Fraser University)

GenerationData SynthesisConvolutional Neural NetworkSupervised Fine-TuningDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkImagePoint CloudMesh

🎯 What it does: A example-driven deep generative network called ShaDDR is constructed, which can refine rough voxel models into high-resolution, textured 3D shapes in less than 1 second.

Shadow Harmonization for Realistic Compositing

Lucas Valença, Jean-Franccois Lalonde

Image HarmonizationRestorationGenerationDiffusion modelGenerative Adversarial NetworkImage

🎯 What it does: Propose a virtual object synthesis method for outdoor scenes that can handle interactions between real and synthetic shadows and generate realistic cast shadows.

ShapeSonic: Sonifying Fingertip Interactions for Non-Visual Virtual Shape Perception

Jialin Huang, Y. Gingold

Audio

🎯 What it does: Developed the ShapeSonic system, which enables non-visual 3D shape perception through audio feedback and tracks the movement of the user's fingertip in three-dimensional space

Shrink & Morph: 3D-Printed Self-Shaping Shells Actuated by a Shape Memory Effect

David Jourdan, Sylvain Lefebvre

OptimizationRobotic IntelligenceDiffusion modelAuto EncoderMesh

🎯 What it does: Propose a computational framework that optimizes the internal structure of a 3D printing plate, enabling it to deform into a free-form surface after heating by utilizing the contraction effect of thermoplastic materials.

SimpleNeRF: Regularizing Sparse Input Neural Radiance Fields with Simpler Solutions

Nagabhushan Somraj (Indian Institute of Science), R. Soundararajan

GenerationData SynthesisDepth EstimationNeural Radiance FieldImage

🎯 What it does: Propose a method that improves geometry and rendering quality under sparse views by training NeRF with two simplified augmentation models (reducing the frequency of positional encoding, removing view-dependent radiance) to supervise the depth of the main model under a few views.

Simultaneous Color Computer Generated Holography

E. Markley, Grace Kuo

GenerationOptimizationConvolutional Neural NetworkTransformerDiffusion modelGenerative Adversarial NetworkContrastive LearningOptical FlowImagePoint CloudPhysics Related

🎯 What it does: Designed and experimentally validated a framework for generating high-quality color holograms using a single SLM and synchronized RGB light source, combining phase models, perceptual loss, and camera calibration.

Single-Image 3D Human Digitization with Shape-guided Diffusion

Badour Albahar, Jia-Bin Huang (University of Maryland)

Image TranslationGenerationPose EstimationDepth EstimationSuper ResolutionConvolutional Neural NetworkTransformerPrompt EngineeringDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderImagePoint CloudMesh

🎯 What it does: Generate fully textured 3D human meshes from a single image.

SinMPI: Novel View Synthesis from a Single Image with Expanded Multiplane Images

Guo Pu (Peking University), Zhouhui Lian (Peking University)

GenerationData SynthesisDepth EstimationTransformerDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningGaussian SplattingImage

🎯 What it does: Propose a single-image view synthesis method called SinMPI, which generates high-quality, freely rotatable 3D-consistent novel views by constructing an extended multi-plane image (MPI).

SLANG.D: Fast, Modular and Differentiable Shader Programming

S. Bangaru, Yong He

OptimizationComputational EfficiencyAuto Encoder

🎯 What it does: Proposes SLANG.D, an extension of the Slang shading language that adds support for first-order automatic differentiation.

Slippage-Preserving Reshaping of Human-Made 3D Content

Chrystiano Araújo, Alla Sheffer

GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkMesh

🎯 What it does: Proposes a reshaping method for non-uniform scaling of artificial 3D models while preserving shear invariance and scale preservation, based on extending the 2D ALUP framework to 3D and incorporating shear protection mechanisms;

SOL-NeRF: Sunlight Modeling for Outdoor Scene Decomposition and Relighting

Jiali Sun, Lin Gao

Neural Radiance FieldGaussian SplattingImageMesh

🎯 What it does: Proposed the SOL-NeRF method, which uses a hybrid illumination representation and signed distance field (SDF) geometry reconstruction to achieve geometric, reflectance, and illumination decomposition and relighting in outdoor scenes.

Sparse Stress Structures from Optimal Geometric Measures

Dylan Rowe, Albert Chern

CompressionOptimization

🎯 What it does: A method is proposed to achieve optimal structural design under given loads by searching for minimal tension structures (tensegrity).

Stable Discrete Bending by Analytic Eigensystem and Adaptive Orthotropic Geometric Stiffness

Zhendong Wang, Huamin Wang

OptimizationDiffusion modelScore-based ModelMeshPhysics Related

🎯 What it does: This paper derives the analytical eigenvalue system of the DAB energy Hessian, utilizes this system to project the indefinite Hessian into a positive semi-definite form, and constructs an orthotropic geometric stiffness matrix with adaptive parameters. It combines the Kirchhoff–Love thin plate theory to adjust the compressive stiffness, in order to address the stability issues caused by the indefiniteness of the DAB model energy and geometric degeneration.