报告人：Dr. Pan Ji
In this talk, I'll present how to formulate convex problems for 3D reconstruction and subspace clustering.
In the first part, we'll revisit the principle of “maximizing rigidity” in structure-from-motion literature, and develop a unified theory which is applicable to both rigid and non-rigid structure reconstruction in a rigidity-agnostic way. We formulate these problems as a convex semi-definite program, imposing constraints that seek to apply the principle of minimizing non-rigidity.
In the second part, I’ll first introduce an important property of subspace clustering, namely subspace self-expressiveness, and discuss how to formulate it as a convex optimization problem. I’ll then talk about how we can approach this problem with deep neural networks in the era of deep learning. Finally, I’ll show how we leverage the subspace constraints to regularize other geometric problems, such as optical flow estimation.
Dr. Pan Ji is currently a researcher at NEC Labs America. Before moving to the U.S. in Feb. 2018, he worked as an ARC Senior Research Associate (Post-Doc) with Pro. Ian Reid at the University of Adelaide since Jul. 2016. Prior to that, he was a PhD student of the Australian National University from Jan. 2013 to Jul. 2016, supervised by Pro. Hongdong Li and Dr. Mathieu Salzmann. His research interests lie in computer vision (especially 3D vision) and unsupervised learning (e.g., clustering). He received the Best Student Paper Award at the International Conference on Image Processing (ICIP) 2014, Paris. From Sep. 2011 to Oct. 2012, Pan was a research assistant in the 3DTV Laboratory of Zhejiang University, China. He received his Bachelor's Degree of Engineering with Chu Kochen Honors from Zhejiang University in Jun. 2011.