SIGGRAPH 2023 Papers — Page 3
ACM SIGGRAPH (Transactions on Graphics) · 212 papers
UniTune: Text-Driven Image Editing by Fine Tuning a Diffusion Model on a Single Image
Dani Valevski (Google Research), Yaniv Leviathan (Google Research)
GenerationTransformerSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: This paper proposes UniTune, a method for text-driven image editing by fine-tuning diffusion models on a single image, enabling diverse edits while preserving the original image's visual and semantic fidelity.
Unsupervised Learning of Robust Spectral Shape Matching
Dongliang Cao (University of Bonn), Florian Bernard (University of Bonn)
Representation LearningGraph Neural NetworkDiffusion modelContrastive LearningPoint CloudMesh
🎯 What it does: Propose an unsupervised spectral shape matching framework that directly learns point-to-point correspondences through deep feature mapping.
UrbanBIS: a Large-scale Benchmark for Fine-grained Urban Building Instance Segmentation
Guoqing Yang, Hui Huang
SegmentationComputational EfficiencyConvolutional Neural NetworkGraph Neural NetworkScore-based ModelContrastive LearningPoint CloudBenchmark
🎯 What it does: This paper proposes UrbanBIS — a large-scale 3D point cloud dataset for urban areas, containing 250 million points, covering an area of 10.78 km², and including fine-grained building category and instance annotations. Based on this dataset, we developed an end-to-end building instance segmentation framework called B-Seg.
Variational quasi-harmonic maps for computing diffeomorphisms
Yu Wang, J. Solomon
Optimization
🎯 What it does: A variational quasi-harmonic mapping is proposed to compute smooth injective (invertible) mappings, and the problem is transformed into an optimal control problem in the parameter space of elliptic partial differential equations. By minimizing a function family, exact injective discrete triangular mesh mappings are obtained.
Variational Shape Reconstruction via Quadric Error Metrics
Tong Zhao, P. Alliez
OptimizationAuto EncoderGenerative Adversarial NetworkContrastive LearningGaussian SplattingPoint CloudMesh
🎯 What it does: Propose a clustering method based on the quaternion error metric to reconstruct the mesh of 3D point clouds.
VideoDoodles: Hand-Drawn Animations on Videos with Scene-Aware Canvases
Emilie Yu, A. Bousseau
Object TrackingGenerationDiffusion modelNeural Radiance FieldOptical FlowVideo
🎯 What it does: Developed an interactive system for inserting hand-drawn animation (video doodles) into videos
ViP-NeRF: Visibility Prior for Sparse Input Neural Radiance Fields
Nagabhushan Somraj (Indian Institute of Science), R. Soundararajan
Depth EstimationNeural Radiance FieldImage
🎯 What it does: A new neural radiance field (NeRF) model called ViP-NeRF is proposed, which introduces a visibility prior to address the training problem under sparse input views.
Virtual Mirrors: Non-Line-of-Sight Imaging Beyond the Third Bounce
Diego Royo (Universidad de Zaragoza), Julio Marco (Universidad de Zaragoza)
Neural Radiance FieldOptical FlowImagePoint CloudComputed TomographyReview/Survey PaperPhysics Related
🎯 What it does: Leveraging the principle of virtual mirrors in the fluctuation domain, this paper proposes using higher-order (fourth and fifth-order) reflected light to achieve non-line-of-sight (NLOS) imaging, addressing the limitations of traditional third-order rebound NLOS imaging, such as the missing cone (null-reconstruction space) and single-corner observation constraints;
Walk on Stars: A Grid-Free Monte Carlo Method for PDEs with Neumann Boundary Conditions
Rohan Sawhney (Carnegie Mellon University), Keenan Crane (Carnegie Mellon University)
OptimizationComputational EfficiencyPoint CloudMeshTabularPhysics Related
🎯 What it does: Proposed a new mesh-free Monte Carlo method called Walk on Stars (WoSt), for solving linear elliptic PDEs (such as the Poisson equation) with mixed Neumann and Dirichlet boundary conditions, enabling direct solutions for arbitrarily complex geometries.
Winding Numbers on Discrete Surfaces
Nicole Feng, Keenan Crane
Mesh
🎯 What it does: Proposes a topologically robust generalization of the winding number on discrete surfaces, capable of generating closed curves, assigning integer labels to regions, and identifying curves that do not enclose any region.
Word-As-Image for Semantic Typography
Shira Iluz, Ariel Shamir (Reichman University)
Image TranslationGenerationTransformerPrompt EngineeringDiffusion modelScore-based ModelImageTextMultimodality
🎯 What it does: Propose a technique for automatically generating word-as-image representations, which can transform the geometric shapes of each letter into graphics that express the semantic meaning of the word while preserving the original font's readability and style.
Zero-shot Image-to-Image Translation
Gaurav Parmar (Carnegie Mellon University), Jun-Yan Zhu (Adobe Research)
Image TranslationGenerationData SynthesisKnowledge DistillationTransformerPrompt EngineeringDiffusion modelScore-based ModelGenerative Adversarial NetworkContrastive LearningImageTextMultimodality
🎯 What it does: Designed a zero-shot image-to-image translation method that requires no training and no manual text prompts, directly utilizing pre-trained text-to-image diffusion models to preserve and edit the structure of real and synthetic images.