Jan 26 (Wed) @ 2:00pm: "Graph Based Methods for Smooth Signal Representation and 3d Point Cloud Compression," Eduardo Pavez, Postdoc Research Assoc, USC
In this talk, I will discuss graph signal processing methods to efficiently represent and process graph structured data. The first part of this talk is dedicated to 3d point clouds, aka volumetric video or holograms, which after images and video, have emerged as the preferred data format used in immersive communications, and AR/VR experiences. Since in recent years, high end phones have started to included depth sensing capabilities, huge amounts of time varying 3d content will be generated and shared over the internet, which will require new compression algorithms. I will discuss some of the challenges arising in 3d point cloud compression, and some graph signal processing tools to address them.
Since the quality of the graph can have a major effect on the effectiveness of graph signal processing algorithms, in the second part of this talk, I will present recent results on graph estimation/learning from data. The proposed methods are based on inverse covariance estimation with graph Laplacian constraints, which lead to compact signal representations. I will discuss various aspects of the graph learning problem including efficient algorithms, and consistency in high dimensions.
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 research is in the areas of graph signal processing, 3d point cloud processing, and compression.
Hosted by: ECE Department
Submitted by: Olivia La Pierre <firstname.lastname@example.org>