On April 11th, 2025 at 2.00 pm the seminar titled "Finding good minimal / optimization problems for structure-from-motion and triangulation" will take place at DEIB Alpha Room (Building 24).

We begin with explaining a new framework for reducing the algebraic complexity of geometric optimization problems, such as triangulation. Our strategy introduces weights to the function to be minimized and finds the best weights such that the number of critical points becomes as small as possible. For instance, we find that 2-view triangulation, which generally has 6 critical points, can be weighted such that it has only 2 critical points that can be found in closed form.
Second, we present a complete catalog of 291 minimal problems for structure-from-motion with uncalibrated cameras observing points, lines and their incidences. We find many new minimal problems with unique solutions that can be found with linear solvers. We compare this catalog with the analog list of 30 minimal problems for calibrated cameras, and - if time permits - special classes of rolling-shutter cameras.
This talk is based on joint works with Albin Ahlbäck, Georg Bökman, Marvin Hahn, Fredrik Kahl, Kim Kiehn, Orlando Marigliano, Tomas Pajdla, Felix Rydell.

This talk is part of the Imaging Seminar from EMJM in Imaging.

Kathlén Kohn received her PhD in Mathematics from TU Berlin in 2018, was a researcher at ICERM (Brown University) and the University of Oslo, and is currently an Associate Professor at KTH Stockholm. She studies the geometry of neural network theory, 3D reconstruction, and maximum likelihood estimation using nonlinear algebra. Her works received a SIAM SIGEST award in 2024, Best Student Paper Award at ICCV 2019, Swedish L’Oréal-Unesco For Women in Science prize 2023. The common thread in her multidiscriplinary research is Metric Algebraic Geometry, on which she co-authored a textbook with the same title. She is a member of the ELLIS society.  https://kathlenkohn.github.io