
Viktor Larsson in currently an Assistant Professor in the division for Computer Vision and Machine Learning at Lund University. Before this he worked as a senior researcher in the Computer Vision and Geometry group at the Institute for Visual Computing at ETH Zurich. His research is on geometric estimation problems in computer vision, with a special focus on problems related to Structure-from-Motion and visual localization.
Talk Title: Structure-from-Motion, from fundamentals to the state-of-the-art
Abstract:
Structure-from-Motion, i.e. jointly estimating camera parameters and a 3D scene representation from images, is a classical problem in computer vision that has been studied for close to half a century. It is now a mature technology that is used in multiple applications, e.g. to calibrate camera geometry or to provide maps for localization. In this talk, I will give an introduction to the underlying principles on which modern frameworks such as COLMAP are built, followed by highlighting some interesting recent work that integrate machine learning into the estimation process.