Schedule

Invited Speakers

Derek Hoiem

UIUC

Derek Hoiem is a Professor at University of Illinois Urbana-Champaign as well as a co-founder and Chief Science Officer of Reconstruct. He is known for his contribution to single-view geometry, 3D scene understanding, and object recognition. His recent research focuses on mixed-view geometry and general purpose vision.

Angjoo Kanazawa

UC Berkeley

Angjoo Kanazawa is an Assistant Professor at UC Berkeley. She also serves on the advisory board of Wonder Dynamics. She aims to build an intelligent system that can capture, perceive, and understand this 4D world from everyday photograph and video.

Marc Pollefeys

ETH Zurich

Marc Pollefeys is a Professor at ETH Zurich. He is known for his work in 3D computer vision and is currently working with a team of scientists and engineers to develop advanced perception capabilities for HoloLens and Mixed Reality. His main area of research is computer vision. One of his main research goals is to develop flexible approaches to capture visual representations of real world objects, scenes and events.

Torsten Sattler

Czech Technical University in Prague

Torsten Sattler is a Senior Researcher at CzechTechnical University in Prague. His research goal is to make 3D computer vision algorithms such as 3D reconstruction and visual localization more robust and reliable through scene understanding while using 3D computer vision methods to train machine learning models.

Siyu Tang

ETH Zurich

Siyu Tang is an Assistant Professor at ETH Zurich. She studies computational models that enable machines to perceive and analyze human motion and activities from visual input.Her gaol is to advance algorithmic foundations of scalable and reliable human digitalization, enabling a broad class of real-​world applications.

Spotlight Speakers / Rising Stars

Yun Chen

University of Toronto

Yun Chen is a Sr. Researcher at Waabi and a Ph.D. student at University of Toronto, working with Raquel Urtasun. His research primarily focuses on scalable neural rendering for real-world applications, particularly in simulation for self-driving vehicles.

Divya Kothandaraman

University of Maryland College Park

Divya Kothandaraman is a Computer Science PhD candidate at the University of Maryland College Park, working with Prof. Dinesh Manocha and Prof. Ming Lin. Her broader research interests lie at the intersection of computer vision, deep learning and multi-modal learning. Her recent works are centered around data efficient personalization and novel-view synthesis.

Yihong Sun

Cornell University

Yihong Sun is a third-year CS PhD Student at Cornell University, advised by Prof. Bharath Hariharan. His work is supported by NSF GRFP and his research interests are computer vision and machine learning, especially in building visual perception systems that can learn from minimal supervision.

Matt Wallingford

University of Washington

Matt Wallingford is a PhD student at the University of Washington advised by Ali Farhadi. His current research focuses on scalable sources of 3D data and computer vision for embodied systems. He aims to build large-scale models of the world with understanding of 3D, temporal dynamics, and objects from visual data.

Related Workshops

• 3D Scene Understanding Workshop covers the broad scope of 3D scene understanding, while our proposed workshop dive deeper into the challenges and methodologies of 3D modeling and reasoning under in-the-wild, challenging, and sometimes extreme setup.
• Long-Term Visual Localization under Changing Conditions focuses on the task of visual localization. Our workshop takes a broader lens, covering the challenges and applications across various 3D computer vision tasks.
• Image Matching: Local Features & Beyond focuses specifically on the task of cross-view feature matching. While 3D modeling and alignment is one of our topics of interest, our focus is broader. Additionally, instead of the traditional, classic setup, we are interested in challenging cases such as sparse viewpoints with small overlaps.
• Learning 3D with Multi-View Supervision explores multi-view cues for 3D deep learning for various tasks, including recognition, detection, segmentation, and generation. While our workshop also concerns these techniques and tasks, we focus more on how to generalize these algorithms to challenging and extreme real-world setup.
• Workshop on Computer Vision in the Wild covers a wide range of topics in computer vision, while we focus on studying the challenges and impact of existing algorithms and systems through the lens of 3D.

Call For Paper

• Submission Portal: OpenReview
• Paper Submission Deadline: August 15 August 23, 2024, 23:59:59 PST
• Notification to Authors: September 4 September 8, 2024
• Camera-ready submission: September 9 September 11, 2024

Topics of Interest

• Alignment: How can we align observations subject to significant variations in appearance, lighting, contents, and viewpoints, and how can we register images with little or no overlap?
• Modeling: How can we construct accurate 3D models from limited, noisy, and incomplete observations?
• Representation: What are the most suitable representations for 3D modeling and reasoning that combine data-driven priors and limited observations? Do we need 3D representations, or would systems for view synthesis suffice?
• Knowledge and Reasoning: How can we incorporate common sense knowledge of 3D objects and scenes, such as part structures, arrangements of objects, physical stability, and affordance, and how can we represent, learn, and encode this knowledge for 3D reasoning?
• Datasets and Benchmarks: What datasets and benchmarks are crucial to validate the effectiveness of 3D algorithms in the wild? And what is currently lacking in this area?
• Risks and ethical considerations: How can we mitigate the risks of these robust 3D modeling and reasoning techniques? How do we address relevant ethical questions, such as invasion of privacy and spreading misinformation.
• Applications: What type of applications can we enable by developing more robust 3D algorithms? What type of modifications should be made? Are there other new and exciting applications for in-the-wild 3D modeling in fields such as construction, agriculture, and remote sensing?

Accepted Papers

• CHOSEN: Contrastive Hypothesis Selection for Multi-View Depth Refinement
• BoostMVSNeRFs: Boosting MVS-based NeRFs to Generalizable View Synthesis in Large-scale Scenes
• SWAG: Splatting in the Wild images with Appearance-conditioned Gaussians
• SpaRP: Fast 3D Object Reconstruction and Pose Estimation from Sparse Views
• Synchronization of Projective Transformations
• CrossScore: Towards Multi-View Image Evaluation and Scoring
• SuperGaussian: Repurposing Video Models for 3D Super Resolution
• SpotLessSplats: Ignoring Distractors in 3D Gaussian Splatting
• HAHA: Highly Articulated Gaussian Human Avatars with Textured Mesh Prior
• SlotLifter: Slot-guided Feature Lifting for Learning Object-centric Radiance Fields
• DecentNeRFs: Decentralized Neural Radiance Fields from Crowdsourced Images
• D$^3$RoMa:Disparity Diffusion-based Depth Sensing for Material-Agnostic Robotic Manipulation
• Enhancing Surface Neural Implicits with Curvature-Guided Sampling and Uncertainty-Augmented Representations
• Deceptive-NeRF/3DGS: Diffusion-Generated Pseudo-Observations for High-Quality Sparse-view 3D Generation
• 3D-GSR: 3D Super-Resolution with Multi-View Adapter and Gaussian Splatting
• latentSplat: Autoencoding Variational Gaussians for Fast Generalizable 3D Reconstruction
• GS2Mesh: Surface Reconstruction from Gaussian Splatting via Novel Stereo Views
• Zero-Shot Metric Depth with a Field-of-View Conditioned Diffusion Model
• WildTalker: Talking Portrait Synthesis In the Wild
• OccFusion: Rendering Occluded Humans with Generative Diffusion Priors
• Free-Viewpoint Video in the Wild Using a Flying Camera
• Dynamic Scene Reconstruction from Single Landscape Image Using 4D Gaussian in the Wild
• PhysPart: Physically Plausible Part Completion for Interactable Objects
• Shape Translation of Dental Point Clouds
• Optimizing Dynamic NeRF and 3DGS with No Video Synchronization
• Coloring the Past: Neural Historical Monuments Reconstruction from Archival Photography
• Single Image, Any Face: Generalisable 3D Face Generation
• G3R: Gradient Guided Generalizable Reconstruction
• Sp2360: Sparse-view 360° Scene Reconstruction using Cascaded 2D Diffusion Priors
• Fast Encoder-Based 3D from Casual Videos via Point Track Processing
• High Frame Rate, Dense Depth Stream by a Low Frame Rate RGB Camera
• TriNeRFLet: A Wavelet Based Triplane NeRF Representation
• PAPR Up-close: Close-up Neural Point Rendering without Holes
• PROSE: Point Rendering of Sparse-Controlled Edits to Static Scenes

Organizers

Wei-Chiu Ma

Cornell

Shenlong Wang

UIUC

Lingjie Liu

UPenn

Yufei Ye

Stanford

Despoina Paschalidou

Nvidia

Natalia Neverova

Meta

David Fouhey

NYU

Shubham Tulsiani

CMU

Qixing Huang

UT Austin

Program Committee