Date＆Time：15:40 – 17:40, December 27, 2021 Location: Duke Ballroom
Intro: The development of machine learning algorithms and 2D/3D sensors have substantially improved the efficiency and quality of 3D content creation. The advantage of data-driven content creation is its ability to learn representations, such as deep neural network, from captured data to boost the performance of 3D content understanding and generation. This symposium will invite three speakers to share their insights and cutting-edge results in learning algorithms for 3D environment creation in AR/VR applications、physics-based motion generation and how the created content can help to improve the accuracy of computer vision tasks.
Organizer: Kun Zhou (Zhejiang university) and Weiwei Xu (Zhejiang University)
Intelligent Medical Information Retrieval Analysis and 3D Reconstruction
The Research on Geometry Analysis and Modeling Based on Deep Geometry Learning
From Sketch to CAD: Sequential CAD Modeling by Sketching in Context
Title: Intelligent Medical Information Retrieval Analysis and 3D Reconstruction
Speaker: Xiaonan Luo，Guilin University of Electronic Technology
Abstract: Intelligent medical information processing technology provides the most convenient way and brand-new experience for digital telemedicine and health-care, as well as medical and health care services for the public; It also provides a new way for the access of public medical information, the construction of information platforms and assisting doctors in diagnosis and treatment. This talk introduces intelligent medical information retrieval and analysis and the theory/method/technology of 3D reconstruction based on the smart home platform. The main contents are as follows: (1) establish an intelligent retrieval and analysis system with massive medical information, which provides a simple, fast, and efficient visualization way for obtaining and analyzing medical health information, and solves the problem of evaluating the utilization level of medical resources for government agencies. (2) develop an intelligent analysis system based on medical images. Through image processing, data mining and other key technologies, it can automatically locate the lesions in medical images and assist doctors to diagnose patients’ conditions, which improves the usage of medical information. (3) research on the dynamic real-time reconstruction of 3D medical data, as well as the on-demand visualization and personalized recommendation of intelligent 3D medical information based on user interaction, so as to improve the accuracy and reliability of computer-aided medical decision-making.
Biography: Xiaonan Luo, is currently a Professor and PhD supervisor with the School of Computer and Information Security, Guilin University of Electronic Technology, Guilin, China. He received the National Science Fund for Distinguished Young Scholars granted by the National Natural Science Foundation of China. He is also one of the directors of Overseas Distinguished Young Scholars. He won the National Science Foundation for Distinguished Young Scholars in 2005. He was selected as the fifth group of Bagui scholars and Guangxi high level talents (Level B) in 2019. He won the second prize of National Science and Technology Progress Award, Guangdong DingYing Science and Technology award, the first prize of Guangdong Science and Technology Progress Award and China Patent Excellence Award, and was approved to receive State Department special allowance. He mainly engages in research and engineering construction of home network, artificial intelligence, graphics and image processing, and 3D Simulation CAD, etc., and is one of the pioneers in the field of home network in China. He has presided over and undertaken more than 40 scientific research projects, such as “Core electronic devices, high-end general chips and basic software products” Major National Science and Technology Projects, National Science Fund for Distinguished Young Scholars, National Natural Science (Key) Fund, etc. He has published more than 100 papers in top international journals and conferences such as ACM/IEEE Transactions、Siggraph, etc. He has been authorized more than 400 national and PCT invention patents and led the formulation of more than 70 national, industrial and local standards.
Title: The Research on Geometry Analysis and Modeling Based on Deep Geometry Learning
Speaker: Lin Gao，Chinese Academy of Sciences
Abstract: Deep geometry learning has become a research hotspot in recent years and is widely used to analyze and model geometric shapes. For the analysis of 3D shapes, we will introduce the point cloud segmentation and recognition network VV-net and the mesh segmentation and recognition network LaplaceNet. We will also introduce the research work of symmetry analysis of 3D models, PRS-Net, which analyzes 3D models through extrinsic symmetry in an unsupervised manner. For 3D deep generative models, we will introduce deep networks SDM-NET which describes and generates geometric details of 3D models through geometric deformation, DSG-NET which encodes and generates complex geometric structure using a recursive graph network, etc. In addition, we will introduce the research works of learning and modeling physical deformation process through deep geometry learning.
Biography: Lin Gao, is an associate professor (PhD supervisor) with Institute of Computing Technology, Chinese Academy of Sciences. His research primarily centers on Intelligent computer graphics and deep geometry learning. He received his PhD degree from Tsinghua University in 2014, and was a visiting professor in RWTH Aachen University at 2016. He has published more than 40 papers in the international journals and conferences such as ACM SIGGRAPH\TOG, IEEE TVCG, IEEE CVPR, ICCV, etc. He has received the awards such as Asia Graphics Young Researcher Award and the First Class of China Computer Federation Technological Invention Award. He has also received funding such as Newton Advanced Fellowship from the Royal Society and Young Elite Scientists Sponsorship Program from Chinese Association for Science and Technology.
Title: From Sketch to CAD: Sequential CAD Modeling by Sketching in Context
Speaker: Hao Pan，Microsoft Research Asia
Abstract: The typical workflow of mechanical product design starts from conception, which is converted to CAD models by mechanical design engineers, analyzed for quality checking and finally sent for production. During conception, designers are used to 2D sketching for the ease of exploration. We notice that the sketching process follows a coarse-to-fine sequence, which largely resembles the modeling command sequence to be issued later in CAD software. Thus, to accelerate the iteration of ideation and design, we propose an automatic translation system that converts sketching sequences into CAD modeling sequences, for intuitive and natural CAD modeling.
The translation needs to resolve inconsistency between two data modalities. On one hand, a sketch is a 2D drawing under a fixed viewpoint, generally rough by hand drawing, and shows a projected image that ambiguously corresponds to multiple 3D shapes. On the other hand, CAD modeling requires precise 3D operations whose results are clear without ambiguity. To bridge the gap, we propose a machine learning model that takes advantage of the contextual information in a modeling sequence to disambiguate the sketch and interpret it into precise CAD modeling operations. In addition, we rely on the compositional property of CAD modeling sequencies to synthesize a large amount of training data that leads to robust and generalizable ML models. Experiments and user evaluations show that our system enables novice users to create diverse and complex CAD models easily after quick learning.
Biography: Hao Pan is a Senior Researcher at Microsoft Research Asia. He obtained his Ph.D. degree from University of Hong Kong in 2015 and joined MSRA after graduation. His research focuses on geometric modeling and processing and 3D vision. He has published more than 10 papers at journals and conferences like ACM TOG, IEEE CVPR and IEEE RAL on topics of curvature guided surface modeling, CAD modeling, mesh model repair, geometric deep learning and object pose estimation, etc. He has served as program committee member for GMP, CVM, 3DV, CAD/Graphics.