Colloquiums and Conferences
A Novel Highway Network for Surface Reconstruction in Computer Graphics
Time: 2026-01-21 15:00:00
Title: A Novel Highway Network for Surface Reconstruction in Computer Graphics
Speaker: Prof. Yaozong Han (Invited by Prof. Qi Ye)
Affiliation: The Chinese University of Hong Kong
Time: January 21, 11:30–12:30
Venue: Room 111, West Building, School of Mathematical Sciences
Speaker Biography:
Yaozong Han is a visiting professor in the Department of Mathematics at The Chinese University of Hong Kong and in the Department of Psychology at the University of Science and Technology of China. He has published more than 173 SCI-indexed papers and has been listed by Professor John Ioannidis's team at Stanford University among the world's top 2% most-cited scientists. His research focuses on computational mathematics and inverse problems in which unknown parameters are inferred from observational data, with applications in engineering, medical imaging, geophysics, non-destructive testing, environmental monitoring and diagnosis. His work typically combines numerical methods for partial differential equations, optimization techniques and statistical modelling to address complex real-world challenges. In 2024 he served as guest editor-in-chief of a special issue on AI-driven technologies in Engineering Analysis with Boundary Elements, and he teaches the MSc course Mathematical Foundations of Artificial Intelligence at The Chinese University of Hong Kong.
Abstract:
Surface reconstruction from point clouds is a fundamental challenge in computer graphics and medical imaging. In this talk, we introduce a novel Square-Highway (SqrHw) Network to explore the application of advanced neural network architectures for an accurate and efficient reconstruction of surfaces from data points. Its performance alongside plain neural networks and a simplified highway network is illustrated through various numerical examples including the reconstruction of simple and complex surfaces, such as spheres, human hands, and intricate models like the Stanford Bunny. We also analyze the impact of factors such as the number of hidden layers, interior and exterior points, and data distribution on surface reconstruction quality. Our results show that the proposed SqrHw architecture outperforms most existing neural network configurations, achieving faster convergence and higher-quality surface reconstructions. Additionally, we demonstrate the SqrHw's ability to predict surfaces over missing data, a valuable feature for challenging applications like medical imaging. Furthermore, our study demonstrates that the proposed SqrHw network yields more stable weight norms and backpropagation gradients compared to the Plain Network architecture. This research not only advances the field of computer graphics but also holds utility for other purposes such as function interpolation and physics-informed neural networks, which integrate multilayer perceptrons into their algorithms.