arXivSub Start free trial

SIGGRAPH 2023 Papers — Page 2

ACM SIGGRAPH (Transactions on Graphics) · 212 papers

Inkjet 4D Print: Self-folding Tessellated Origami Objects by Inkjet UV Printing

Koya Narumi, Yoshihiro Kawahara

Diffusion modelScore-based ModelOptical FlowMeshPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Proposed the Inkjet 4D Print method, which realizes self-foldable 3D origami tessellations by printing 2D patterns on both sides of a thermally shrinkable substrate using a commercial inkjet UV printer.

Interactive Hair Simulation on the GPU using ADMM

G. Daviet

OptimizationComputational EfficiencyPoint CloudMeshPhysics Related

🎯 What it does: A local-global solver specifically designed for simulating discrete elastic rods (Discrete Elastic Rods, DER) with Coulomb friction has been developed, making full use of the parallel computing capabilities of modern GPUs.

Inverse Global Illumination using a Neural Radiometric Prior

Saeed Hadadan (University of Maryland), Matthias Zwicker (University of Maryland)

OptimizationComputational EfficiencyRepresentation LearningDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderImageBenchmark

🎯 What it does: This paper proposes a method for inverse rendering using neural radiance fields and radiometric priors, leveraging the residual of the rendering equation as regularization to capture global illumination without explicitly tracing multiple light paths;

Iterative α -(de)Blending: a Minimalist Deterministic Diffusion Model

E. Heitz, T. Chambon

GenerationConvolutional Neural NetworkTransformerDiffusion modelScore-based ModelImageStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Propose an iterative method based on α-mixing/de-mixing (IADB), achieving a deterministic mapping from arbitrary noise distributions to target distributions, similar to diffusion models;

Juxtaform: interactive visual summarization for exploratory shape design

Karran Pandey, Karan Singh

GenerationOptimizationPrompt EngineeringVision Language ModelDiffusion modelMesh

🎯 What it does: Propose an interactive visual summarization method called Juxtaform for exploring, analyzing, selecting, and improving large shape collections to support shape design workflows.

Kernel-Based Frame Interpolation for Spatio-Temporally Adaptive Rendering

Karlis Martins Briedis, Christopher Schroers

RestorationGenerationComputational EfficiencyGaussian SplattingOptical FlowImageVideo

🎯 What it does: Propose a frame interpolation method based on spatially varying kernels, and introduce a content-adaptive spatiotemporal strategy to reduce rendering costs.

Key-Locked Rank One Editing for Text-to-Image Personalization

Yoad Tewel (NVIDIA), Y. Atzmon

GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelContrastive LearningImageTextMultimodality

🎯 What it does: Propose a text-to-image (T2I) personalization method called Perfusion, which utilizes cross-attention key-locked rank-1 editing to achieve high-fidelity, controllable generation of user-provided visual concepts, while supporting the combination of multiple concepts in the same image.

Language-based Photo Color Adjustment for Graphic Designs

Zhenwei Wang (City University of Hong Kong), Rynson W. H. Lau

Image TranslationImage HarmonizationGenerationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkTransformerVision Language ModelDiffusion modelGenerative Adversarial NetworkContrastive LearningImageTextMultimodality

🎯 What it does: This paper proposes a language-based photo color adjustment method (LangRecol), which can automatically predict source colors based on user-provided multi-grained natural language instructions and perform natural color remapping in local regions of illustrations.

LatentAvatar: Learning Latent Expression Code for Expressive Neural Head Avatar

Yuelang Xu (Tsinghua University), Yebin Liu (Tsinghua University)

GenerationRepresentation LearningConvolutional Neural NetworkTransformerNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningImageVideo

🎯 What it does: Propose LatentAvatar, which learns and reproduces neural head avatars driven by latent expression codes.

Learning Physically Simulated Tennis Skills from Broadcast Videos

Haotian Zhang, Kayvon Fatahalian

Robotic IntelligenceTransformerReinforcement LearningVideoPhysics Related

🎯 What it does: A system is proposed that learns diverse physical simulation tennis skills by leveraging a large-scale tennis demonstration extracted from broadcast videos. The system employs a hierarchical model combining low-level imitation strategies with high-level motion planning strategies to achieve coherent rally play with various types of shots, spins, and styles in a physics simulation environment.

Listen, Denoise, Action! Audio-Driven Motion Synthesis with Diffusion Models

Simon Alexanderson (KTH Royal Institute of Technology), G. Henter

GenerationData SynthesisPose EstimationTransformerPrompt EngineeringMixture of ExpertsDiffusion modelScore-based ModelVideoSequentialAudio

🎯 What it does: This paper proposes an audio-driven human motion synthesis method based on diffusion models, which can generate dance and gestures synchronized with music or speech, and supports style control and style interpolation.

Local Deformation for Interactive Shape Editing

Honglin Chen, Kevin Wampler

OptimizationComputational EfficiencyPoint CloudMeshGraph

🎯 What it does: A new SC-L1 regularization method is proposed for achieving local shape editing based on elastic energy, along with a three-block ADMM optimization framework that enables real-time interactive performance.

Locally Attentional SDF Diffusion for Controllable 3D Shape Generation

Xin Zheng (Tsinghua University), H. Shum (Tsinghua University)

GenerationData SynthesisConvolutional Neural NetworkTransformerVision Language ModelDiffusion modelScore-based ModelAuto EncoderImageMesh

🎯 What it does: Propose a 3D shape generation framework LAS-Diffusion based on two-stage diffusion, achieving controllable generation from sketches through local attention.

Locally Meshable Frame Fields

Heng Liu, D. Bommes

OptimizationMeshBenchmark

🎯 What it does: Study the topology of frame fields, derive their meshability conditions, and propose an algorithm to automatically convert non-meshable frame fields into locally meshable frame fields.

Masonry Shell Structures with Discrete Equivalence Classes

Rulin Chen, Ying He

Optimization

🎯 What it does: A method is proposed to model masonry shell structures as discrete equivalence classes, and through hierarchical clustering and optimization, reusable shell templates are generated to achieve approximable and manufacturable seamless structures for free-form surfaces.

Materialistic: Selecting Similar Materials in Images

Prafull Sharma (MIT), V. Deschaintre

SegmentationConvolutional Neural NetworkTransformerSupervised Fine-TuningAuto EncoderContrastive LearningImage

🎯 What it does: Propose a material selection method based on user guidance, which can automatically identify and segment all pixel regions with the same material as the query pixel in natural images;

MERF: Memory-Efficient Radiance Fields for Real-time View Synthesis in Unbounded Scenes

C. Reiser, Peter Hedman

GenerationData SynthesisComputational EfficiencyDiffusion modelNeural Radiance FieldAuto EncoderGaussian SplattingImageMesh

🎯 What it does: Propose MERF — a memory-efficient radiance field representation that enables real-time view synthesis of large-scale unbounded scenes in the browser.

Mesh Density Adaptation for Template-based Shape Reconstruction

Yucheol Jung, Seungyong Lee

OptimizationGraph Neural NetworkDiffusion modelAuto EncoderContrastive LearningPoint CloudMesh

🎯 What it does: This paper proposes a three-dimensional shape reconstruction method based on template mesh deformation, which uses grid density adaptive energy to densify vertices in complex structures, thereby improving reconstruction accuracy in inverse rendering and non-rigid registration.

Meso-Facets for Goniochromatic 3D Printing

Lubna Abu Rmaileh, Alan Brunton

GenerationOptimizationComputational EfficiencyDiffusion modelScore-based ModelAuto EncoderGenerative Adversarial NetworkMesh

🎯 What it does: Proposes controlling multi-material 3D printers by adding meso-facets to the input surface to achieve viewpoint-dependent color effects on arbitrary surfaces

MesoGen: Designing Procedural On-Surface Stranded Mesostructures

Élie Michel, T. Boubekeur

GenerationData SynthesisDiffusion modelAuto EncoderGenerative Adversarial NetworkMesh

🎯 What it does: Proposes MesoGen, a block-based authoring tool for designing three-dimensional tessellated structures with non-periodic self-similarity on macro surfaces, supporting real-time editing while ensuring structural continuity.

Metameric: Spectral Uplifting via Controllable Color Constraints

M. van de Ruit, E. Eisemann

Image TranslationImage HarmonizationRestorationImage

🎯 What it does: Proposes a controllable spectral lifting method that allows users to define texture appearance under different illumination conditions, and determines the achievable color variation space through constrained optimization, supporting interactive adjustment;

Micro-Mesh Construction

A. Maggiordomo, M. Tarini

CompressionAuto EncoderGaussian SplattingOptical FlowMesh

🎯 What it does: This paper proposes an automated method to convert existing millions of triangle meshes into the μ-mesh compressed format, fully utilizing the self-boundary, compressed displacement mapping scheme supported by native GPU hardware ray-tracing, thereby enhancing geometric details without significantly increasing memory and runtime costs.

Microfacet Theory for Non-Uniform Heightfields

Eugene d'Eon, Tizian Zeltner

🎯 What it does: Proposes a new NDF combination method for microsurface theory in non-uniform elevation fields, derives the corresponding importance sampling scheme, and extends the Smith model to support piecewise constant NDFs and material properties that vary with height, while verifying the accuracy through Monte Carlo simulation.

Min-Deviation-Flow in Bi-directed Graphs for T-Mesh Quantization

Martin Heistermann, D. Bommes

OptimizationFlow-based ModelMeshGraph

🎯 What it does: Proposed the Bi-MDF (Bidirectional Network Minimum Deviation Flow) problem to model and efficiently solve the T-Mesh quantization problem, developed a fast approximate solver and an iterative refinement algorithm based on graph matching, which can obtain the exact solution of Bi-MDF.

Modulating Pretrained Diffusion Models for Multimodal Image Synthesis

Cusuh Ham (Georgia Institute of Technology), T. Hinz (Adobe Research)

GenerationData SynthesisConvolutional Neural NetworkTransformerPrompt EngineeringDiffusion modelScore-based ModelImageTextMultimodality

🎯 What it does: This paper proposes a lightweight multimodal modulation module (MCM), which achieves multimodal control over images by modulating the noise prediction during sampling, while keeping the parameters of the pre-trained diffusion model unchanged.

MoiréTag: Angular Measurement and Tracking with a Passive Marker

Simeng Qiu, W. Heidrich

Object TrackingPose EstimationComputational EfficiencyDiffusion modelScore-based ModelOptical FlowImage

🎯 What it does: A snapshot-based method for angle measurement and tracking using a passive moiré marker generated by a double-sided binary structure on a glass plate is proposed, and complete 6D tracking and camera intrinsic parameter estimation are achieved through a chirp model.

Motion from Shape Change

Oliver Gross, P. Schröder

OptimizationOptical FlowMeshPhysics RelatedOrdinary Differential Equation

🎯 What it does: Propose an algorithm that takes a sequence of shapes changing over time as input, and uses shape geometry and variational principles to output the corresponding motion in world coordinates;

Multi-Layer Thick Shells

Yunuo Chen, Minchen Li

OptimizationMesh

🎯 What it does: A grid-based multilayer thick shell dynamics simulation method is proposed, which can accurately capture the constitutive behavior in the thickness direction;

NeRF-Texture: Texture Synthesis with Neural Radiance Fields

Yihua Huang, Lin Gao

GenerationData SynthesisNeural Radiance FieldAuto EncoderContrastive LearningImage

🎯 What it does: Proposes a texture synthesis method based on NeRF, which can capture and synthesize textures with three-dimensional geometric structures from multi-view images.

Nerfstudio: A Modular Framework for Neural Radiance Field Development

Matthew Tancik (University of California, Berkeley), Angjoo Kanazawa (University of California, Berkeley)

Data SynthesisOptimizationComputational EfficiencyNeural Radiance FieldImageVideo

🎯 What it does: Developed a modular and user-friendly Python framework called Nerfstudio for NeRF research and practical applications.

NeRO: Neural Geometry and BRDF Reconstruction of Reflective Objects from Multiview Images

Yuan Liu (University of Hong Kong), Wenping Wang (Texas A&M University)

RestorationDiffusion modelNeural Radiance FieldAuto EncoderImage

🎯 What it does: Propose the NeRO method, which uses multi-view RGB images to simultaneously reconstruct the geometry and BRDF of reflective objects under unknown lighting and without object masks through neural rendering.

NeRSemble: Multi-view Radiance Field Reconstruction of Human Heads

Tobias Kirschstein (Technical University of Munich), M. Nießner

RestorationGenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingOptical FlowImageVideo

🎯 What it does: Constructed a large-scale multi-view facial video dataset (4,734 clips, 222 people, 31.7M frames), and proposed the NeRSemble method, which uses deformation fields and a multi-resolution hash grid ensemble to perform high-fidelity reconstruction and novel view rendering of dynamic faces.

Neural Biplane Representation for BTF Rendering and Acquisition

Jiahui Fan, Ling-Qi Yan

CompressionRepresentation LearningDiffusion modelNeural Radiance FieldAuto EncoderImage

🎯 What it does: A dual-plane representation BTF model is proposed, which achieves fast compression and inference by utilizing feature textures in the semi-vector field and spatial domain, along with a small universal decoder, and can reconstruct a complete BTF from about 20 photos taken with a smartphone camera and flash.

Neural Face Rigging for Animating and Retargeting Facial Meshes in the Wild

Dafei Qin (University of Hong Kong), T. Komura

RestorationGenerationPose EstimationConvolutional Neural NetworkGraph Neural NetworkSupervised Fine-TuningDiffusion modelAuto EncoderContrastive LearningImagePoint CloudMesh

🎯 What it does: Developed an end-to-end deep learning framework called Neural Face Rigging (NFR), which can automatically bind and reposition 3D facial meshes of arbitrary topology, unknown identities, and expressions.

Neural Parametric Mixtures for Path Guiding

Honghao Dong (Peking University), Sheng Li (Peking University)

OptimizationComputational EfficiencyTransformerDiffusion modelScore-based ModelFlow-based ModelNeural Radiance FieldAuto EncoderGaussian SplattingImageBenchmark

🎯 What it does: Developed a parameterized hybrid model based on neural implicit representations (Neural Parametric Mixtures) for path guiding, which can continuously encode the spatial-direction distribution and decode it into a vMF mixture via a lightweight MLP, enabling efficient sampling.

Neural Partitioning Pyramids for Denoising Monte Carlo Renderings

Martin Bálint, Rafał K. Mantiuk

RestorationConvolutional Neural NetworkRecurrent Neural NetworkDiffusion modelAuto EncoderImage

🎯 What it does: Propose a new low-pass pyramid filter kernel architecture to achieve high-quality denoising in rendering with low sample counts;

Neural Prefiltering for Correlation-Aware Levels of Detail

Philippe Weier, Philipp Slusallek

CompressionComputational EfficiencyRepresentation LearningDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderPoint CloudMesh

🎯 What it does: Proposed a practical general neural appearance filtering pipeline for physically based rendering, which can aggregate visibility of multi-level details based only on local information and directly learn a compressed representation of light transport inside voxels.

Neural Progressive Meshes

Yun-Chun Chen (University of Toronto), Alec Jacobson (Adobe Research)

CompressionConvolutional Neural NetworkDiffusion modelAuto EncoderContrastive LearningMesh

🎯 What it does: Propose a sub-resolution based encoder-decoder network to achieve progressive compression and reconstruction of 3D meshes; the server maps high-resolution meshes to sparse face-level features, and the client instantaneously generates low-resolution meshes through a pre-trained sub-resolution decoder, and can further enhance quality at any time by transmitting additional features.

Neural Volumetric Reconstruction for Coherent Synthetic Aperture Sonar

Albert W. Reed (Arizona State University), Suren Jayasuriya (Arizona State University)

OptimizationDiffusion modelNeural Radiance FieldAuto EncoderUltrasoundAudio

🎯 What it does: Proposes a neural volume reconstruction method based on analysis-synthesis optimization for coherent reconstruction of synthetic aperture sonar (SAS) scenes.

NeuSample: Importance Sampling for Neural Materials

Bing Xu, R. Ramamoorthi

Flow-based Model

🎯 What it does: Evaluated and compared multiple PDF learning methods for sampling spatially varying neural materials, and proposed a new method variant.

NOFA: NeRF-based One-shot Facial Avatar Reconstruction

Wang-Wang Yu, Baoyuan Wu (Chinese University of Hong Kong)

RestorationGenerationPose EstimationDepth EstimationSuper ResolutionConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingImageVideoMesh

🎯 What it does: Propose a one-shot facial avatar reconstruction framework based on NeRF called NOFA, which can generate high-quality 3D facial avatars using a single image and supports fine-grained expression-driven facial reproduction.

OctFormer: Octree-based Transformers for 3D Point Clouds

Peng-Shuai Wang

Object DetectionSegmentationTransformerAuto EncoderContrastive LearningPoint Cloud

🎯 What it does: Propose OctFormer, a Transformer framework based on octrees, which achieves efficient linear complexity learning for large-scale 3D point clouds through Octree Attention;

Orientable Dense Cyclic Infill for Anisotropic Appearance Fabrication

X. Chermain, S. Lefebvre

Optimization

🎯 What it does: Proposes a method to generate surfaces with adjustable directional optical roughness (mimicking the effect of brushed steel) using standard fused filament fabrication (FFF) 3D printers

P2M: A Fast Solver for Querying Distance from Point to Mesh Surface

Chen Zong (Shandong University), Changhe Tu (Shandong University)

OptimizationComputational EfficiencyPoint CloudMesh

🎯 What it does: This paper proposes the P2M algorithm, which preprocesses vertices using a KD-tree and constructs an interception table (recording the interception relationships between vertices and edges/faces), thereby first finding the nearest vertex during queries and then quickly locating the nearest geometric primitive through the interception table;

Parameter-space ReSTIR for Differentiable and Inverse Rendering

Wesley Chang, Tzu-Mao Li

OptimizationComputational EfficiencyDiffusion modelScore-based ModelNeural Radiance Field

🎯 What it does: Proposes an algorithm to reuse Monte Carlo gradient samples in inverse rendering to improve gradient estimation accuracy and accelerate convergence

Patternshop: Editing Point Patterns by Image Manipulation

Xingchang Huang (Max-Planck-Institut fur Informatik), G. Singh

GenerationData SynthesisDiffusion modelAuto EncoderContrastive LearningGaussian SplattingOptical FlowImagePoint Cloud

🎯 What it does: A low-dimensional perceptual embedding space was constructed, mapping the density and correlation of point patterns into a three-channel image, enabling intuitive editing through image editing software;

PCBend: Light Up Your 3D Shapes With Foldable Circuit Boards

Marco Freire, S. Lefebvre

OptimizationDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningMesh

🎯 What it does: A low-cost surface display technology was developed, which attaches a flexible PCB to a 3D printed support structure, covering the surface and arranging individually addressable RGB LEDs.

Perspective-Correct VR Passthrough Without Reprojection

Grace Kuo, N. Matsuda

Optical Flow

🎯 What it does: Designed and implemented a computational camera capable of directly sampling the light in front of the user's eyes to improve VR passthrough effects.

PhotoMat: A Material Generator Learned from Single Flash Photos

Xilong Zhou (Texas A&M University), N. Kalantari

GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkImage

🎯 What it does: Train a material generation model (PhotoMat) based solely on real smartphone flash photos, capable of outputting SVBRDF parameters usable in 3D rendering.

PMP: Learning to Physically Interact with Environments using Part-wise Motion Priors

Jinseok Bae (Seoul National University), Y. Kim

Adversarial AttackRobotic IntelligenceReinforcement LearningDiffusion modelGenerative Adversarial NetworkVideoSequential

🎯 What it does: This study proposes a method that utilizes part-wise motion priors (PMP) to train full-body characters to perform complex interactions and movements in physical simulations by combining motion capture samples from individual body parts;

Polynomial 2D Green Coordinates for Polygonal Cages

Élie Michel, Jean-Marc Thiery

Mesh

🎯 What it does: Proposes conformal polynomial coordinates for closed polygonal cages to define 2D deformation fields from sparse control points

PolyStokes: A Polynomial Model Reduction Method for Viscous Fluid Simulation

Jonathan Panuelos, Christopher Batty

OptimizationComputational EfficiencyPhysics Related

🎯 What it does: Proposed a reduced-order fluid model utilizing incompressible polynomial vector fields to accelerate the solution of unsteady Stokes problems.

PoseVocab: Learning Joint-structured Pose Embeddings for Human Avatar Modeling

Zhe Li (Tsinghua University), Yebin Liu (Tsinghua University)

GenerationPose EstimationGraph Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderVideoMesh

🎯 What it does: Propose PoseVocab, a joint structured pose embedding method, used to learn high-frequency details of human animated avatars;

Potentially Visible Hidden-Volume Rendering for Multi-View Warping

Jang-Hoon Kim, Sungkil Lee

Computational EfficiencyOptical Flow

🎯 What it does: The model and rendering algorithm for potential visible hidden volumes (PVHVs) were studied and implemented, aimed at improving rendering efficiency and quality in multi-view image warping.

Progressive null-tracking for volumetric rendering

Zackary Misso, Wojciech Jarosz

Neural Radiance Field

🎯 What it does: Proposes two new techniques to improve the null-collision method used in programmatic media with unknown boundary attenuation values, enhancing the robustness and efficiency of rendering.

Pyramid Texture Filtering

Qing Zhang (Sun Yat-sen University), Weihua Zheng

RestorationGaussian SplattingImage

🎯 What it does: A texture smoothing method based on the Gaussian-Laplacian pyramid is proposed, achieving texture removal while preserving structural details through progressive upsampling at the coarsest level.

QuestEnvSim: Environment-Aware Simulated Motion Tracking from Sparse Sensors

Sunmin Lee (Seoul National University), Alexander W. Winkler

Pose EstimationReinforcement LearningPoint CloudTabularTime Series

🎯 What it does: Propose a system named QuestEnvSim, which generates complete body poses and achieves natural interaction with environmental objects by utilizing sparse sensor inputs that only include the helmet and hand controller poses and scene information, through physical simulation and deep reinforcement learning.

Random-Access Neural Compression of Material Textures

Karthik Vaidyanathan (NVIDIA), Aaron E. Lefohn

CompressionAuto EncoderImage

🎯 What it does: Propose a random access neural compression method tailored for material textures, which can decompress high-resolution textures at low bitrates while maintaining high quality.

Real-Time Radiance Fields for Single-Image Portrait View Synthesis

Alex Trevithick (University of California San Diego), Koki Nagano (NVIDIA)

GenerationData SynthesisComputational EfficiencyKnowledge DistillationRepresentation LearningConvolutional Neural NetworkDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkImagePoint CloudMesh

🎯 What it does: Propose an encoder that can instantly and in real-time infer a 3D renderable tri-plane representation directly from a single portrait or cat face image.

Recursive Control Variates for Inverse Rendering

Baptiste Nicolet, T. Müller

OptimizationComputational EfficiencyPhysics Related

🎯 What it does: Propose a recursive control variates method to reduce variance and bias in physics-based differentiable rendering (PBDR).

Relighting Neural Radiance Fields with Shadow and Highlight Hints

Chong Zeng (Zhejiang University), Xin Tong (Microsoft Research Asia)

RestorationGenerationNeural Radiance FieldImage

🎯 What it does: Train a neural implicit radiance field based on SDF using a small number of unstructured photos to achieve relighting from arbitrary viewpoints

Revisiting controlled mixture sampling for rendering applications

Qingqin Hua, Philipp Slusallek

OptimizationMixture of Experts

🎯 What it does: This paper improves and extends the control variate theory proposed in previous work, making it applicable to scenarios involving millions of integrals in practical rendering, and achieves robust hybrid sampling and near-optimal MIS weighting for common applications such as light source selection, BSDF sampling, and path guiding.

Rhizomorph: The Coordinated Function of Shoots and Roots

Bosheng Li, Wojciech Pałubicki

Diffusion modelAgriculture RelatedPhysics Related

🎯 What it does: Proposed a physically feasible soil model, a root development process model adaptable to different soils, and introduced remote signaling to coordinate the growth of branches and roots, ultimately achieving a unified modeling of trees and their root systems.

Robust Low-Poly Meshing for General 3D Models

Zhen Chen, Xifeng Gao

Diffusion modelAuto EncoderGenerative Adversarial NetworkMesh

🎯 What it does: A robust remeshing method is proposed, which can automatically generate visually faithful meshes with a low number of polygons.

RSMT: Real-time Stylized Motion Transition for Characters

Xiangjun Tang (Zhejiang University), Xiaogang Jin (Zhejiang University)

GenerationPose EstimationRecurrent Neural NetworkTransformerMixture of ExpertsAuto EncoderGenerative Adversarial NetworkContrastive LearningOptical FlowVideoSequential

🎯 What it does: This paper proposes a method for real-time stylized intermediate motion synthesis (RSMT), which can generate natural, high-quality, and controllable transition motions given the start and end frames, time duration, and target style.

Sag-Free Initialization for Strand-Based Hybrid Hair Simulation

J. Hsu, Kui Wu

OptimizationPhysics Related

🎯 What it does: Proposed a four-stage framework without vertical initialization for solving the stable quasi-static configuration of a hybrid Lagrangian/Eulerian chain hair dynamic system.

ScanBot: Autonomous Reconstruction via Deep Reinforcement Learning

Hezhi Cao, Ligang Liu

Robotic IntelligenceRecurrent Neural NetworkTransformerReinforcement LearningPoint CloudMesh

🎯 What it does: Proposed a reconstruction-oriented automatic scanning method called ScanBot, which utilizes hierarchical deep reinforcement learning for global ROI planning and local NBV planning to improve scanning efficiency and reconstruction quality.

Scratch-based Reflection Art via Differentiable Rendering

Pengfei Shen, Ligang Liu

OptimizationDiffusion modelNeural Radiance FieldGenerative Adversarial NetworkMesh

🎯 What it does: Propose a three-dimensional reflective art based on scratches, using multi-view optimization and differentiable rendering methods, utilizing analytical scratch rendering to complete the reflector design and achieve virtual presentation and physical carving within a few minutes

Second-order Stencil Descent for Interior-point Hyperelasticity

L. Lan, Yin Yang

OptimizationComputational EfficiencyPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Implemented a finite element hyperelastic simulation algorithm based on the internal point method on the GPU

Seeing Photons in Color

Sizhuo Ma, Mohit Gupta

RestorationImagePhysics Related

🎯 What it does: Proposed a computational photography technique for reconstructing high-quality color images from binary frames captured by SPAD arrays

Semantics and Scheduling for Machine Knitting Compilers

Jenny Lin, J. Mccann

Optimization

🎯 What it does: Formal semantics of the knitout language for knitting machines are proposed, and program equivalence is defined using the theory of caging knots. A series of rewrite rules are proven correct, and it is demonstrated how these rules can serve as the basis for compilers and optimizers, ensuring that the same knitted objects are generated on target machines.

Semi-supervised reference-based sketch extraction using a contrastive learning framework

Chang Wook Seo (KAIST), Jun-yong Noh

Image TranslationGenerationDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningImageMultimodality

🎯 What it does: Designed a semi-supervised reference-based multi-modal sketch extraction network that can learn to imitate the style of reference sketches and generate high-quality sketches without paired data.

ShapeCoder: Discovering Abstractions for Visual Programs from Unstructured Primitives

R. K. Jones (Brown University), Daniel Ritchie (Brown University)

Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningPoint CloudMeshChain-of-Thought

🎯 What it does: This paper proposes ShapeCoder, which can automatically learn interpretable visualization programs and their abstract functions from unstructured geometric primitive datasets, and use these abstracts to generate more compact and structurally reasonable programs;

Shortest Path to Boundary for Self-Intersecting Meshes

Heng Chen (University of Utah), C. Yuksel (University of Utah)

OptimizationComputational EfficiencyRobotic IntelligenceDiffusion modelScore-based ModelOptical FlowPoint CloudMeshGraphRetrieval-Augmented GenerationChain-of-ThoughtStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: A precise shortest path computation method from internal points to the boundary in self-intersecting meshes is proposed, and it is applied for efficient self-collision handling.

Simulation and Retargeting of Complex Multi-Character Interactions

Yunbo Zhang (Georgia Institute of Technology), Jungdam Won (Seoul National University)

Graph Neural NetworkReinforcement LearningVideoGraphSequential

🎯 What it does: Train physical simulation characters using deep reinforcement learning to reproduce and redirect complex interaction actions among multiple characters.

Single Image Neural Material Relighting

J. Bieron, P. Peers

Image TranslationRestorationGenerationDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkImage

🎯 What it does: Developed a neural material relighting method for re-visualizing photos of planar spatially varying materials under new viewpoints and illumination conditions.

Single-Shot Implicit Morphable Faces with Consistent Texture Parameterization

Connor Z. Lin (Stanford University), S. Khamis

RestorationGenerationPose EstimationDepth EstimationTransformerDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderContrastive LearningImageMesh

🎯 What it does: This paper proposes a single-image reversible implicit deformable face model that can achieve editable and animatable 3D facial avatars while preserving high-quality geometry and texture.

Sketch-Guided Text-to-Image Diffusion Models

Andrey Voynov (Google Research), D. Cohen-Or

GenerationConvolutional Neural NetworkTransformerPrompt EngineeringDiffusion modelScore-based ModelImageText

🎯 What it does: Propose a method that leverages the inference process of a pre-trained text-to-image diffusion model, combining sketches as spatial guidance to achieve the conversion from sketches to high-quality images.

SketchFaceNeRF: Sketch-based Facial Generation and Editing in Neural Radiance Fields

Lin Gao, Hongbo Fu

GenerationData SynthesisDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkImage

🎯 What it does: Proposes the SketchFaceNeRF method, achieving the generation and editing of realistic 3D facial NeRF models from 2D sketches.

Skin-Screen: A Computational Fabrication Framework for Color Tattoos

Michal Piovarči, B. Bickel

GenerationData SynthesisDiffusion modelAuto EncoderGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes a computational manufacturing framework for tattoos by constructing an automated tattooing device, preparing realistic skin-colored silicone plates, and establishing a tattoo appearance prediction model.

Somigliana Coordinates: an elasticity-derived approach for cage deformation

Jiong Chen, M. Desbrun

Diffusion modelScore-based ModelOptical FlowMeshPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: A cage deformer based on Somigliana coordinates was developed, which avoids common issues such as shear artifacts and insufficient volume control in traditional cage deformers by utilizing linear elasticity theory.

Spectral Coarsening with Hodge Laplacians

A. Keros

Mesh

🎯 What it does: Propose a novel spectral-level simplification method based on the Hodge Laplacian operator, which can perform controllable feature-preserving simplification on triangle meshes, tetrahedral meshes, and simplicial complexes.

Split-Lohmann Multifocal Displays

Yingsi Qin, Aswin C. Sankaranarayanan

Optical FlowPhysics Related

🎯 What it does: Designed a multifocal display system capable of creating a dense focal plane stack in a single shot, called Split‑Lohmann multifocal display.

Stealth Shaper: Reflectivity Optimization as Surface Stylization

Kenji Tojo (University of Tokyo), Nobuyuki Umetani (University of Tokyo)

OptimizationDiffusion modelAuto EncoderContrastive LearningOptical FlowPoint CloudMeshPhysics Related

🎯 What it does: By optimizing the surface normals on a triangular mesh, it is possible to significantly reduce or adjust the reflectance of a model while preserving its original geometry, achieving invisibility or other optical functions.

StripMaker: Perception-driven Learned Vector Sketch Consolidation

Chenxi Liu, A. Sheffer

ClassificationRecognitionTransformerSupervised Fine-TuningContrastive LearningImage

🎯 What it does: Propose a learning-based vector sketch automatic merging method called StripMaker, which utilizes two perception-oriented classifiers to achieve automatic merging of multiple overlapping strokes in sketches.

StyleAvatar: Real-time Photo-realistic Portrait Avatar from a Single Video

Lizhen Wang (Tsinghua University), Yebin Liu (Tsinghua University)

Image TranslationRestorationGenerationPose EstimationConvolutional Neural NetworkDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningOptical FlowImageVideo

🎯 What it does: Propose StyleAvatar, a real-time high-fidelity full-body portrait video avatar reconstruction and reenactment method based on StyleGAN, which can achieve high-quality portrait reconstruction within a single video segment.

Sum-of-Squares Collision Detection for Curved Shapes and Paths

Paul Zhang, Rasmus Tamstorf

OptimizationComputational EfficiencyRobotic IntelligenceMesh

🎯 What it does: Proposes techniques to reduce the computational cost of Sum-of-Squares Programming (SOSP), and utilizes these improvements to increase the speed of Bézier triangle collision detection by up to 300×, while implementing SOSP-based collision detection on conical three-dimensional cylinders, and provides algebraic formulas for rigid body motion that simultaneously handle surface geometry and trajectories.

Surface and Edge Detection for Primitive Fitting of Point Clouds

Yuanqi Li, Yanwen Guo

RecognitionPose EstimationDepth EstimationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderContrastive LearningPoint Cloud

🎯 What it does: Proposes a surface and edge detection network called SED-Net for fitting point cloud geometric primitives, which can simultaneously detect parametric surfaces (including B-spline patches) and edges;

Surface Simplification using Intrinsic Error Metrics

Hsueh-Ti Derek Liu (Roblox), Keenan Crane (Carnegie Mellon University)

OptimizationDiffusion modelScore-based ModelAuto EncoderGenerative Adversarial NetworkContrastive LearningMesh

🎯 What it does: This paper proposes an intrinsic triangulation simplification method for solving equations on surfaces, which maintains the approximate intrinsic geometry through local vertex removal, reconstruction, and intrinsic Delaunay flips, generating a multi-resolution hierarchy directly usable for numerical solutions.

Synthesizing Dexterous Nonprehensile Pregrasp for Ungraspable Objects

Sirui Chen (University of Hong Kong), C.Karen Liu

OptimizationRobotic IntelligenceGraph Neural NetworkReinforcement LearningDiffusion modelAuto EncoderContrastive LearningOptical FlowPoint CloudMeshGraph

🎯 What it does: The study proposes a framework based on graph search, optimal control, and a learning objective function to synthesize non-grasping pregrasp motions that enable originally ungraspable objects to become graspable under environmental constraints.

Synthesizing Physical Character-Scene Interactions

Mohamed Hassan (Electronic Arts), Xue Bin Peng (NVIDIA)

Robotic IntelligenceReinforcement LearningGenerative Adversarial NetworkPoint CloudMeshSequentialPhysics Related

🎯 What it does: Trained physics-simulated characters demonstrate natural and realistic performance in three scene interaction tasks (sitting, lying down, and carrying objects).

Temporal Set Inversion for Animated Implicits

Kavosh Jazar, P. Kry

OptimizationComputational EfficiencyMesh

🎯 What it does: Proposes an algorithm that utilizes temporal consistency to perform local re-computation on implicit surfaces for closed-loop animation, in order to maintain global error invariant over time and save resources.

TextDeformer: Geometry Manipulation using Text Guidance

William Gao, Rana Hanocka

GenerationOptimizationGraph Neural NetworkTransformerVision Language ModelDiffusion modelAuto EncoderContrastive LearningImageTextMesh

🎯 What it does: Automatically deform input 3D meshes based on given text prompts, making them conform to the text description while preserving the original semantic correspondence.

TEXTure: Text-Guided Texturing of 3D Shapes

Elad Richardson, D. Cohen-Or

GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelImageTextMesh

🎯 What it does: Proposes a text-guided 3D shape texture generation, editing, and transfer method based on diffusion models; achieves high-quality textures by iteratively rendering, painting, and projecting back to texture maps from different perspectives;

The Visual Language of Fabrics

V. Deschaintre, B. Masiá

GenerationRetrievalRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodalityRetrieval-Augmented Generation

🎯 What it does: This paper creates a dataset called text2fabric, which includes 45,000 rendered images of 3,000 fabric materials and more than 15,000 natural language descriptions, and performs lexical, attribute, and structural analysis on these descriptions; subsequently, the dataset is used to fine-tune CLIP and BLIP for fine-grained text retrieval, image retrieval, and fabric description generation.

Topology driven approximation to rational surface-surface intersection via interval algebraic topology analysis

Jin-San Cheng, Ming Li

OptimizationComputational EfficiencyRobotic Intelligence

🎯 What it does: A topology-driven hybrid symbolic-numerical framework based on interval algebraic topology analysis (IATA) is proposed for approximately solving surface-surface intersections (SSI) of rational parametric surfaces.

Toward Optimized VR/AR Ergonomics: Modeling and Predicting User Neck Muscle Contraction

Yunxiang Zhang (New York University), Qi Sun (New York University)

Human-Computer InteractionPose EstimationOptimizationExplainability and InterpretabilityComputational EfficiencyReinforcement Learning from Human FeedbackConvolutional Neural NetworkTabularTime SeriesSequentialBiomedical Data

🎯 What it does: This study collects synchronized data of neck muscle contraction levels (MCL) and head motion (posture and acceleration) by using surface electromyography (EMG) sensors on VR headset wearers. Based on this data, a biophysical model (MCLNet) is constructed to estimate the MCL of completed head movements. Subsequently, TrajectoryNet is used to regress typical motion trajectories between initial and end postures, further predicting the MCL before the head movement begins, thereby assessing the user's potential neck discomfort in advance.

Towards Attention–aware Foveated Rendering

Brooke Krajancich (Stanford University), Gordon Wetzstein (Stanford University)

CompressionComputational EfficiencyContrastive LearningImageVideo

🎯 What it does: This paper proposes and verifies a perspective-dependent attention perception model through experimental measurements of how attention allocation affects visual sensitivity, and integrates it into a gaze-related rendering pipeline, achieving more adaptive gaze-focused rendering.

Towards Material Digitization with a Dual-scale Optical System

Elena Garces, Jorge López-Moreno

Image TranslationData SynthesisDiffusion modelOptical FlowImagePoint Cloud

🎯 What it does: A dual-scale optical system was developed to digitize material surface properties at the microscale and mesoscale, including geometric features, anisotropic reflectance, and transmittance, with microscale information propagated to the mesoscale through neural networks.

Trim Regions for Online Computation of From-Region Potentially Visible Sets

Philip Voglreiter, D. Schmalstieg

Computational EfficiencyOptical FlowMesh

🎯 What it does: This paper proposes the Trim Regions method, which generates visibility sets (PVS) for arbitrary scenes under real-time conditions by performing controlled erosion of object silhouettes in image space.