Special RI Seminar Please Note Different Day and Time

Abstract:  At first sight, 3D reconstruction can be considered a solved problem. The principles are well understood and we can reconstruct a wide range of objects and scenes using active as well as passive reconstruction approached. However, most of these reconstructions are not convincing when really scrutinized by a human observer. In this talk, I will give an overview over our work on 3D reconstruction with a particular focus on reconstructing thin surfaces. I will then introduce a novel holistic evaluation metric for complete reconstruction and rendering pipelines that focuses on the resulting image quality and tries to take human observations — whether it “looks right” — into account.

Bio: 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. After receiving his Ph.D. from the Max Planck-Institute for Computer Science and Saarland University in Saarbrücken, Germany, he spent two years as a Feodor Lynen fellow of the Alexander von Humboldt Foundation at the University of Washington in Seattle, WA, USA before he moved to TU Darmstadt. One focus of his research is on 3D scene reconstruction from general input data – such as images from online photo collections – and photo-realistic rendering approaches built on such reconstructed models.
Host: Yaser Sheikh

Point of Contact: Stephanie Matvey (smatvey@andrew.cmu.edu)