CS Graduate Seminar | On solving large-scale MRFs on GPUs and on evaluating image-based modeling and rendering systems

Sep 26 2016 12:00 PM - Sep 26 2016 01:00 PM

By Professor Michael Goesele (TU Darmstadt, Germany)

Image-based modeling and rendering systems have made tremendous progress in recent years. Motivated by our own modeling and rendering pipeline, I will present two separate contributions:First, I will discuss the challenges in evaluating the end-to-end quality of image-based modeling and rendering system. I will then introduce our proposed methodology, discuss several quality metrics and show its relationship to existing benchmarks. One of the key advantages of this approach is that it does not require ground truth geometry. Finally, I will conclude the first part of my talk by introducing our new image-based modeling and rendering benchmark. In the second part of the talk, I will focus on solving large-scale MRFs, one of the key bottlenecks in the texturing stage of our reconstruction pipeline. I will first discuss strategies to parallelize the solution of large scale MRFs with a focus on massively-parallel approaches suitable for modern GPUs. I will then introduce our solver that supports arbitrary MRF topologies efficiently and can handle arbitrary, dense or sparse label sets as well as label cost functions. Together with two additional heuristics for further acceleration, our solver performs favorably even compared to modern specialized solvers in terms of speed and solution quality, especially when solving very large MRFs. This is joint work with Michael Waechter, Mate Beljan, Simon Fuhrmann, Nils Moehrle, Johannes Kopf, Daniel Thuerck, Sven Widmer, Max von Buelow, Patrick Seemann and Marc E. Pfetsch.

Biography: Dr.-Ing. Michael Goesele is a professor in TU Darmstadt’s Department of Computer Science where he heads the research group “Graphics, Capture and Massively Parallel Computing”. He received his diploma in Computer Science from Ulm University in 1999. He then joined the Max Planck Institute for Computer Science in Saarbrücken and earned his Ph.D. from Saarland University in 2004. After a two year postdoctoral stay at the University of Washington (Seattle), he joined TU Darmstadt in 2007. His research interests include computer graphics, computer vision, and massively parallel computing. 

Learn More

More Information:

​Contact person for more info: Prof. Wolfgang Heidrich; email: wolfgang.heidrich@kaust.edu.sa

Date: Monday 26th September 2016

Time: 12:00pm - 1:00pm

Location: Building 9, Lecture Hall I Room 2322

Brown-bag lunch will be available at 11:45 am