Information Theory Forum: Graph based compression of 3D point cloud attributes
Zoom only
Abstract: 3D point clouds have become the preferred data format to represent 3D objects, scenes and people. Driven by immersive communications, XR applications, and autonomous vehicles, compression of 3D point clouds has become an active research area. Unlike images and video where pixels lie on a 2D grid, because of the irregular object geometry, 3D point clouds occupy a sparse subset of the 3D grid. As a consequence, several techniques essential for efficient video compression such as: block transforms, filters, and spatial and temporal prediction, become much more challenging while having increased computational complexity. In this talk, I will present our recent work on transforms, intra prediction and filtering for compression of 3D point cloud attributes using tools from graph signal processing. I will show that a proposed multi-resolution graph transform, combined with multi-resolution intra prediction can achieve state of the art attribute coding performance.
Bio: Eduardo Pavez received the B.S. and M.Sc. degrees in electrical engineering from the University of Chile, Santiago, Chile, in 2011 and 2013, respectively, and the Ph.D. degree in electrical engineering from the University of Southern California, in 2019. He was an intern at Microsoft Research, and Mitsubishi Electric Research Laboratories, in 2016 and 2017, respectively. He is currently a Post-Doctoral Research Associate at the University of Southern California. His work on point cloud and video compression received the ICIP 2020 best paper award (1st place) and ICIP 2022 best paper award (3rd place). His research is in the areas of graph signal processing, 3D point clouds, and compression.