CNCC2024
Brief introduction of the forum:
At the end of the 3D reconstruction is Gauss? Advances in the construction and mapping of three-dimensional Gaussian expressions
Time: 13:30-17:30, October 24
Location: Hall 2, 1st Floor, Summer Garden-Haiyan Hall
Note: If there is any change, please refer to the final information on the official website (https://ccf.org.cn/cncc2024).
Gaussian expression has become the most high-profile research direction in the field of 3D scene reconstruction and drawing. In just one year, the quality and performance of Gaussian expression in 3D reconstruction and drawing have been fully proven in academia and industry, and have begun to be applied in important research fields such as SLAM and digital human. At the same time, a large number of research works are being carried out and published, including but not limited to further improving Gaussian expressions to improve their drawing speed, quality and accuracy of geometric representations, as well as the application of Gaussian expressions in high-fidelity digital human modeling and driving, real-time 3D reconstruction, reverse rendering, large-scale scene reconstruction and other aspects. In this context, this forum intends to share reports and discuss from multiple dimensions such as the theory, technology and application of Gaussian expressions, so as to further promote the development of 3D reconstruction theory and technology based on Gaussian expressions.
Forum Agenda
| order | topic | Keynote speaker | unit |
| 1 | 3D Gaussian reconstruction and generation of dynamic scenes | Liu Yebin | Tsinghua University |
| 2 | Real-time Gaussian SLAM: Real-time high-fidelity 3D reconstruction of large indoor scenes | Shao Tianjia | Zhejiang University |
| 3 | Reconstruction, editing and generation of digital humans based on Gaussian expressions | Zhang Juyong | University of Science and Technology of China |
| 4 | Geometrically accurate 3D reconstruction based on 2DGS and its application | Goldman Sachs | The University of Hong Kong |
| 5 | City-level scene reconstruction and rendering for embodied intelligent simulation | Deb | Shanghai Artificial Intelligence Innovation Center |
| Panel link | Forum Speaker |
Introduction of the chairman and guests of the forum
Chair of the Forum
Zhou Kun
Director of CCF Computer Aided Design and Graphics Committee, Professor of Zhejiang University, Director of CAD&CG National Key Laboratory
Professor of School of Computer Science, Zhejiang University, ACM/IEEE Fellow. He received his Ph.D. degree in engineering from Zhejiang University in 2002, was selected as a Changjiang Scholar Distinguished Professor by the Ministry of Education in 2007, and was awarded the National Science Fund for Distinguished Young Scholars in 2008. His research interests include computer graphics, computer vision, human-computer interaction, and virtual reality. He has published more than 100 papers in ACM/IEEE Transactions and obtained more than 80 invention patents. He has won 2 second prizes of the National Natural Science Award, Chen Kah Kee Young Scientist Award, Science Exploration Award, ACM SIGGRAPH Test-of-Time Award, MIT TR35 and other domestic and foreign awards.
Forum Speaker
Liu Yebin
He is a tenured professor at Tsinghua University
Winner of the National Outstanding Youth Fund. His research interests are 3D vision, digital human reconstruction and generation. He is the deputy director of the 3D Vision Committee of the Chinese Society of Image and Graphics. He won the first prize of the National Technological Invention Award in 2012 (ranked 3rd) and the first prize of the Technological Invention Award of the Chinese Institute of Electronics in 2019 (ranked 1st).
Title: 3D Gaussian Reconstruction and Generation of Dynamic Scenes
Abstract:3D Gaussian point cloud rendering (Guassian Splatting) has emerged as a new differentiable rendering technology. Compared with NeRF, it has gained a lot of attention in recent years due to its higher rendering quality, faster rendering speed, and more compatibility with traditional rendering pipelines. With the advantages of Guassian Splatting in 3D scene expression and rendering, this paper introduces the research work of the reporter in dynamic 3D reconstruction, dynamic 3D editing, and digital human generation. The report will also discuss the future development trend of 3D representation based on the current video generation model.
Shao Tianjia
Researcher at Zhejiang University
Winner of the National Excellent Youth Fund. He was a lecturer at the University of Leeds, United Kingdom. His research interests include 3D scene reconstruction, digital human modeling, physical simulation, etc. He serves as a member of the ACM SIGGRAPH 2022/2023 Program Committee, co-chair of the CVM 2022 Program Committee, and Secretary General of the Virtual Reality Branch of the Chinese Institute of Electronics. He won the second prize of the 2020 National Natural Science Award (ranked 2nd).
Title: Real-time Gaussian SLAM: Real-time high-fidelity 3D reconstruction of large indoor scenes
Abstract:Real-time reconstruction of 3D scenes is a key supporting technology in the fields of augmented reality, unmanned driving, and low-altitude economy. The traditional reconstruction technology has high geometric accuracy, but cannot present a high-fidelity appearance. The NeRF-based reconstruction technology has high computational cost and high memory consumption, making it difficult to achieve real-time reconstruction. In this report, I will introduce the preliminary exploration of the research group in the direction of real-time Gaussian SLAM. We propose the first real-time Gaussian SLAM system, which simultaneously obtains high-quality geometry, high-fidelity appearance, and accurate camera pose during the scanning process. The system reconstruction performance reaches real-time, the memory consumption is low, and it supports real-time high-fidelity 3D reconstruction of large-scale indoor scenes.
Zhang Juyong
Professor, University of Science and Technology of China
He was funded by the Outstanding Youth Fund of the National Foundation of China and the Outstanding Member of the Youth Promotion Association of the Chinese Academy of Sciences. He received his bachelor's degree in computer science from USTC in 2006 and his Ph.D. degree from Nanyang Technological University in Singapore in 2011, and worked as a postdoctoral researcher at the Switzerland Federal Institute of Technology Lausanne from 2011 to 2012. His research interests are computer graphics and 3D vision, and his recent research interests are the efficient and high-fidelity 3D digitization of the real physical world based on neural implicit representation, reverse rendering and numerical optimization methods, as well as the creation of high-fidelity virtual digital content.
Title: High-Fidelity Virtual Human: From Modeling, Driving to Generation and Editing
Abstract:In recent years, the representation methods represented by NeRF and 3D Gaussian have made great breakthroughs in the reconstruction of people, objects and scenes due to their strong fitting expression ability and differentiability. In addition to modeling, the multimodal drive of digital humans, the editing and control of appearance and emotion are equally important. In this report, I will introduce the research work of the research group in the rapid reconstruction of virtual digital humans, physics-based movement and clothing drives, and multimodal digital human editing.
Goldman Sachs
Associate Professor, The University of Hong Kong
He has received support from the Overseas High-level Talent Youth Program and the Shanghai Outstanding Academic Leader Program. His research interests include 3D reconstruction, image and video understanding and generation, 3D generation, AI4Science, etc. He has served as a field chair for more than a dozen top conferences (CVPR, NeurIPS, ICCV, ACM MM, ECCV, etc.) and as a publicity chair for CVPR 2024. He also serves as an associate editor of IEEE TPAMI, TMM, TCSVT, among others.
Title: Geometrically accurate 3D reconstruction based on 2DGS and its application
Abstract:Three-dimensional Gaussian splash (3DGS) has made revolutionary progress in 3D reconstruction, achieving high-quality new perspective synthesis and fast rendering speed. However, due to the multi-view inconsistencies of the 3D Gaussian scatters, 3DGS cannot accurately represent the surface. We propose a novel method based on 2D Gaussian splash (2DGS) to model and reconstruct geometrically accurate optical radiation fields from multi-view images. At our core, we depicted 3D objects with a set of 2D directional plane Gaussian disks and introduced a perspecular-accurate 2D splash process. We have proven the effectiveness of 2DGS in several applications such as 4D reconstruction, human reconstruction, and scene editing.
Deb
Young Scientist of Shanghai Artificial Intelligence Laboratory
Part-time doctoral supervisor at Shanghai Jiao Tong University and Fudan University. His research interests include generative artificial intelligence and its applications, and he has published more than 60 related papers, including AnimateDif and the world's first city-level real-life 3D large-scale model scholar skyline series.
Title: City-level Scene Reconstruction and Rendering for Embodied Intelligent Simulation
Abstract:Recently, 3D-GS has shown unique advantages over NeRF in terms of rendering speed, quality, and compatibility with traditional pipelines. In this report, the presenter will discuss the possibilities and limitations of using 3D-GS to build large-scale virtual simulation environments, and introduce the team's exploration of the limitations of solving the excessive number of model parameters and the lack of geometric accuracy, including Scaffold-GS, Octree-GS, and GSDF. At the same time, the team's progress in the 3D-GS supporting system will also be introduced.
About CNCC2024
CNCC2024 will be held on October 24-26 in Hengdian Town, Dongyang City, Zhejiang Province, with the theme of "Developing New Quality Productivity, Computing Leads the Future". The three-day conference included 18 invited reports, 3 conference forums, 138 thematic forums, 34 thematic activities and more than 100 exhibitions. More than 800 speakers, including Turing Award winners, academicians of the Chinese Academy of Sciences and the Chinese Academy of Sciences, top scholars at home and abroad, and well-known entrepreneurs, looked forward to cutting-edge trends and shared their innovative achievements. More than 10,000 people are expected to attend.