Viktor Larsson

Computer Vision Researcher

viktor.jpg

I am an Assistant Professor at the Centre for Mathematical Sciences at Lund University working in Computer Vision and Machine Learning. Previously I worked as a PostDoc and senior researcher at the Computer Vision and Geometry group lead by Marc Pollefeys at ETH Zurich

My research is mainly focused on robust estimation problems that appear in 3D computer vision (e.g. Structure-from-Motion, visual localization, SLAM, and dense geometry estimation). Currently I am exploring how to best integrate deep learning with classical geometry-based pipelines. In particular, identifying and replacing hand-crafted heuristics or implicit assumptions with more data-driven alternatives.

I am looking for talented students (PhD/MSc) who are interested in doing exciting research in computer vision, geometry and machine learning.

news

Mar 05, 2024 Two papers accepted for CVPR 2023. Open PhD position in my group, check it out here!
Jul 18, 2023 Four papers accepted for ICCV 2023, including our paper on joint point+line matching GlueStick. More details about the other papers will follow soon :)
Jul 13, 2023 I gave an invited plenary talk at SIAM AG 2023 conference about the intersection between computer vision and algebraic geometry.
Apr 01, 2023 Five papers accepted for CVPR 2023. Our paper on line mapping LIMAP was selected for highlight (top 2.5% submissions)!
Jul 01, 2022 One paper and one tutorial accepted for ECCV 2022. More details here. You can also find a short version of the tutorial here from a recent talk in the Learning Machines seminar series at RI.SE.
Apr 01, 2022 Started as Assistant Professor at Lund University, Sweden. I will also serve as Area Chair for 3DV 2022.
Mar 03, 2022 Three papers accepted for CVPR 2022.
Oct 01, 2021 Our paper Pixel-Perfect Structure-from-Motion with Featuremetric Refinement won the Best Student Paper award at ICCV 2021. You can check it out here.
Jul 01, 2021 Three papers accepted for ICCV 2021.
Mar 01, 2021 Four papers accepted for CVPR 2021.