Members of Michael Kazhdan’s research group will present two papers at the Special Interest Group in Computer Graphics and Interactive Techniques (SIGGRAPH) 2022 Conference, to be held August 8-11 in Vancouver, Canada.
The paper “Möbius Convolutions for Spherical CNNs” details the processing of spherical images using geographical formulas, proposing the first convolutional neural networks (CNNs) that are adaptable and flexible when responding to various viewpoints of spherical images. This results in faster rendering of spherical images and more utilitarian methods for their development. Along with Kazhdan, the study’s authors include doctoral student Thomas W. Mitchell; and Noam Aigerman and Vladimir G. Kim, research scientists at Adobe Research.
In the second paper, titled “Variational Quadratic Shape Functions for Polygons and Polyhedra,” Kazhdan and colleagues focus on the development of formulas to accurately render geometric meshes for faster computation speeds. Together with collaborators from TU Dortmund University and ETH Zurich, he has helped develop a two-level system to solve partial differential equations (PDEs) that are essential to the execution of computer graphics. The team’s method is not only more speed efficient in loading and producing these polygonal and polyhedral meshes, but also cost efficient.
Michael Kazhdan is a professor of computer science and an industry leader in image and geometry processing. He is an affiliate of the JHU Institute for Data-Intensive Engineering and Science (IDIES) and is a 2016 recipient of the National Science Foundation’s Early CAREER Award.
For more information on SIGGRAPH 2022, click here.