arXivSub Start free trial

SIGGRAPH Asia 2023 Papers — Page 3

ACM SIGGRAPH Asia (Transactions on Graphics) · 224 papers

Subspace Mixed Finite Elements for Real-Time Heterogeneous Elastodynamics

Ty Trusty, D. Levin

Diffusion modelScore-based ModelOptical FlowMeshPhysics RelatedStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: Developed a reduced-order hybrid finite element solver for real-time heterogeneous elastic dynamics, capable of maintaining rich rotational motion with a low number of iterations;

Subspace-Preconditioned GPU Projective Dynamics with Contact for Cloth Simulation

Xuan Li, Chenfanfu Jiang

Optimization

🎯 What it does: Propose an efficient fabric simulation method that combines subspace integration with parallel iterative relaxation, capable of simultaneously efficiently computing low-frequency and high-frequency motions;

Text-Guided Synthesis of Eulerian Cinemagraphs

Aniruddha Mahapatra (Carnegie Mellon University), Junchen Zhu

GenerationData SynthesisConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelDiffusion modelGaussian SplattingOptical FlowImageVideoTextMultimodality

🎯 What it does: Propose a method to automatically generate cinemagraphs (semi-static and semi-dynamic images) from text descriptions, capable of generating continuous looping animations of fluid elements (such as waterfalls, rivers, and clouds).

Text-Guided Vector Graphics Customization

Peiying Zhang, Jing Liao

GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageTextMesh

🎯 What it does: This paper proposes a complete pipeline for automatically generating customized SVGs based on a single SVG example and text prompts.

Texture Atlas Compression Based on Repeated Content Removal

Yuzhe Luo, Xifeng Gao

CompressionImageMesh

🎯 What it does: Propose a content-aware, automated loss compression pipeline for texture atlases, including image segmentation, remeshing, UV unwrapping, and texture baking;

The effect of display capabilities on the gloss consistency between real and virtual objects

Bin Chen, Rafał K. Mantiuk

🎯 What it does: Achieve perceptual gloss matching between virtual and real objects on an HDR 3D augmented reality display through an accurate imaging pipeline, and conduct multi-factor gloss matching experiments, revealing the impact of ability on gloss consistency.

The Shortest Route is Not Always the Fastest: Probability-Modeled Stereoscopic Eye Movement Completion Time in VR

Budmonde Duinkharjav, Qi Sun

Explainability and InterpretabilityComputational EfficiencyRepresentation LearningGaussian SplattingTabularTime SeriesBiomedical DataStochastic Differential Equation

🎯 What it does: Collected 12,672 trials in virtual reality to build a probabilistic model for predicting the completion time of panoramic eye movements.

Thin On-Sensor Nanophotonic Array Cameras

Praneeth Chakravarthula (Princeton University), Felix Heide (Princeton University)

RestorationDiffusion modelOptical FlowImagePhysics Related

🎯 What it does: Propose an ultra-thin sensor-level nanophotonic array camera, which directly deposits multiple miniature metal lenses on the sensor cover glass, achieving high-resolution imaging with a 100° wide field of view at a thickness of 700 nm and a focal length of 2.5 mm, and image restoration is realized through differentiable optical design and probabilistic diffusion models.

Topology Guaranteed B-Spline Surface/Surface Intersection

Jieyin Yang, Dong-ming Yan

Optimization

🎯 What it does: A B-spline surface intersection algorithm that ensures topological correctness is proposed, capable of handling complex topologies such as multiple branches, intersecting singularities, isolated singularities, high-order contacts, and boundary intersections.

ToRoS: A Topology Optimization Approach for Designing Robotic Skins

Juan Montes Maestre, Bernhard Thomaszewski

OptimizationRobotic IntelligenceDiffusion modelScore-based Model

🎯 What it does: Proposes a method for automatically generating 3D printed reinforcement patterns in soft robotic skin design through tri-domain topology optimization, used to control silicone actuators to achieve a variety of large-range deformations

Towards Garment Sewing Pattern Reconstruction from a Single Image

Lijuan Liu (Sea AI Lab), Shuicheng Yan (Sea AI Lab)

Image TranslationRestorationPose EstimationDepth EstimationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkContrastive LearningImagePoint CloudMesh

🎯 What it does: This paper proposes a method for reconstructing garment sewing patterns from a single RGB image, capable of recovering complete garment sewing diagrams from ordinary camera photographs.

Towards Practical Capture of High-Fidelity Relightable Avatars

Haotian Yang (Kuaishou Technology China), Chongyang Ma (Kuaishou Technology China)

GenerationComputational EfficiencyRepresentation LearningTransformerDiffusion modelNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingImageVideoMesh

🎯 What it does: Proposed the TRAvatar framework, which can capture and reconstruct high-fidelity, relightable 3D facial avatars in real-time from dynamic image sequences captured under multi-view photogrammetry, without the need for tracking, and supports video-driven animation and environmental relighting.

Transparent Object Reconstruction via Implicit Differentiable Refraction Rendering

Fangzhou Gao, Jiawan Zhang

RestorationDiffusion modelNeural Radiance FieldImage

🎯 What it does: Reconstructing the geometry of transparent objects in unknown natural scenes without human intervention

TwinTex: Geometry-Aware Texture Generation for Abstracted 3D Architectural Models

Weidan Xiong (Shenzhen University), Hui Huang (Shenzhen University)

RestorationGenerationTransformerDiffusion modelAuto EncoderContrastive LearningOptical FlowImageMesh

🎯 What it does: Proposes an全自动 texture mapping framework called TwinTex, which can generate high-quality, geometry-aware texture maps for abstracted 3D building models.

UVDoc: Neural Grid-based Document Unwarping

Floor Verhoeven (ETH Zurich), O. Sorkine-Hornung

RestorationConvolutional Neural NetworkDiffusion modelScore-based ModelFlow-based ModelRectified FlowNeural Radiance FieldAuto EncoderContrastive LearningImageMeshBenchmark

🎯 What it does: Propose a single-image document de-warping method based on a dual-headed fully convolutional network, utilizing 3D meshes and 2D inverse mappings to jointly predict and achieve document de-warping.

Variational Barycentric Coordinates

Ana Dodik (MIT), Justin Solomon (MIT)

OptimizationDiffusion modelScore-based ModelNeural Radiance FieldAuto EncoderContrastive LearningOptical FlowMeshStochastic Differential Equation

🎯 What it does: Proposes a method that uses neural fields to variational optimize the Barycentric coordinates of points inside polygonal/polyhedral cages, allowing the coordinates to automatically adapt according to any energy function (such as TV, ARAP, Dirichlet, etc.).

VASCO: Volume and Surface Co-Decomposition for Hybrid Manufacturing

Fanchao Zhong, Lin Lu

OptimizationPoint CloudMesh

🎯 What it does: A computational framework is proposed to optimize the sequence of additive and subtractive processes while ensuring tool accessibility. The VASCO problem is defined, the hybrid-fabricability geometric attribute is introduced, and a beam-guided top-down block decomposition algorithm is designed. The framework is applied to a 5-axis hybrid manufacturing platform and evaluated on various three-dimensional shapes for process planning.

VET: Visual Error Tomography for Point Cloud Completion and High-Quality Neural Rendering

Linus Franke (Friedrich-Alexander-Universität Erlangen-Nürnberg), Marc Stamminger (Friedrich-Alexander-Universität Erlangen-Nürnberg)

RestorationGenerationSuper ResolutionComputational EfficiencyConvolutional Neural NetworkTransformerDiffusion modelNeural Radiance FieldAuto EncoderContrastive LearningGaussian SplattingImageVideoPoint Cloud

🎯 What it does: Proposes a point cloud completion and neural rendering framework based on Visual Error Tomography (VET), which can automatically clean and complete sparse or missing geometric structures.

ViCMA: Visual Control of Multibody Animations

Doug L. James, D. Levin

Optimization

🎯 What it does: Propose a multi-body animation method that utilizes object motion and visibility for visual control, applicable to large-scale, contact-rich rigid and deformable body examples, with an overall cost comparable to a single simulation.

VMesh: Hybrid Volume-Mesh Representation for Efficient View Synthesis

Yuanchen Guo, Songiie Zhang

Computational EfficiencyRepresentation LearningDiffusion modelNeural Radiance FieldContrastive LearningGaussian SplattingOptical FlowImagePoint CloudMesh

🎯 What it does: Propose a hybrid voxel-mesh representation called VMesh, which uses a triangle mesh to capture large-scale geometric structures, while using sparse voxels to compensate for the fine details that meshes struggle to represent, achieving high-quality real-time view synthesis.

VR-NeRF: High-Fidelity Virtualized Walkable Spaces

Linning Xu (Chinese University of Hong Kong), Christian Richardt (Meta)

GenerationData SynthesisOptimizationComputational EfficiencyNeural Radiance FieldImagePoint Cloud

🎯 What it does: A complete system for capturing, reconstructing, and real-time VR rendering of high-fidelity walkable spaces (VR-NeRF) was proposed, enabling immersive roaming at 36-72 FPS on devices such as Quest Pro.

Warped-Area Reparameterization of Differential Path Integrals

Peiyu Xu, Shuang Zhao

OptimizationComputational EfficiencyNeural Radiance FieldPhysics RelatedStochastic Differential Equation

🎯 What it does: Proposes a mathematical formulation for reparameterization in the path space, avoiding explicit sampling of geometric boundaries, thereby achieving more efficient differentiable rendering.

What is the Best Automated Metric for Text to Motion Generation?

Jordan Voas (University of Texas at Austin), Raymond Mooney (University of Texas at Austin)

GenerationData SynthesisPose EstimationTransformerSupervised Fine-TuningVision-Language-Action ModelContrastive LearningTextMultimodality

🎯 What it does: This paper systematically evaluates various automated evaluation metrics for the task of generating skeletal motions from natural language, and proposes a novel multimodal BERT-style evaluation model called MoBERT.

Zero-Shot 3D Shape Correspondence

Ahmed Abdelreheem (King Abdullah University of Science and Technology), Peter Wonka (King Abdullah University of Science and Technology)

ClassificationRecognitionSegmentationTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextPoint CloudMeshBenchmarkChain-of-Thought

🎯 What it does: Propose a fully zero-shot 3D shape correspondence method, which utilizes a language-visual foundation model to classify shapes, generate semantic regions, perform zero-shot semantic segmentation, and obtain fine-grained correspondence through functional mapping.