Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Active Exploration for Sparse Visual Localization

Lidholm, Johanna LU ; Dillén, Ludvig LU orcid ; Kukelova, Zuzana ; Sattler, Torsten and Larsson, Viktor LU orcid (2026) p.338-347
Abstract
In this paper, we present a pipeline for visual localization using 2D-3D matching, yielding high localization accuracy comparable with hierarchical methods based on 2D-2D matching. This is accomplished through an active exploration strategy, which adaptively explores multiple parts of the scene to find the region that generates the most inlier matches, followed by refining the pose with a pose-to-pixel uncertainty re-matching scheme. To justify our design choices, we conduct extensive ablations demonstrating clear improvements over related approaches. Our method achieves competitive accuracy on the large-scale benchmarks Aachen Day & Night, RobotCar, CMU Seasons, and Gangnam Station, while maintaining a low runtime. The code can be... (More)
In this paper, we present a pipeline for visual localization using 2D-3D matching, yielding high localization accuracy comparable with hierarchical methods based on 2D-2D matching. This is accomplished through an active exploration strategy, which adaptively explores multiple parts of the scene to find the region that generates the most inlier matches, followed by refining the pose with a pose-to-pixel uncertainty re-matching scheme. To justify our design choices, we conduct extensive ablations demonstrating clear improvements over related approaches. Our method achieves competitive accuracy on the large-scale benchmarks Aachen Day & Night, RobotCar, CMU Seasons, and Gangnam Station, while maintaining a low runtime. The code can be found at https://github.com/jojjo99/active_exploration. (Less)
Please use this url to cite or link to this publication:
author
; ; ; and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
published
subject
keywords
Localization, 2D-3D matching, Image Retrieval
host publication
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings
pages
10 pages
publisher
IEEE Computer Society
language
English
LU publication?
yes
id
7978aa9d-21d8-44d6-b304-adf39982b356
alternative location
https://openaccess.thecvf.com/content/CVPR2026F/papers/Lidholm_Active_Exploration_for_Sparse_Visual_Localization_CVPRF_2026_paper.pdf
date added to LUP
2026-06-26 12:48:13
date last changed
2026-09-23 13:12:18
@inproceedings{7978aa9d-21d8-44d6-b304-adf39982b356,
  abstract     = {{In this paper, we present a pipeline for visual localization using 2D-3D matching, yielding high localization accuracy comparable with hierarchical methods based on 2D-2D matching. This is accomplished through an active exploration strategy, which adaptively explores multiple parts of the scene to find the region that generates the most inlier matches, followed by refining the pose with a pose-to-pixel uncertainty re-matching scheme. To justify our design choices, we conduct extensive ablations demonstrating clear improvements over related approaches. Our method achieves competitive accuracy on the large-scale benchmarks Aachen Day & Night, RobotCar, CMU Seasons, and Gangnam Station, while maintaining a low runtime. The code can be found at https://github.com/jojjo99/active_exploration.}},
  author       = {{Lidholm, Johanna and Dillén, Ludvig and Kukelova, Zuzana and Sattler, Torsten and Larsson, Viktor}},
  booktitle    = {{Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings}},
  keywords     = {{Localization; 2D-3D matching; Image Retrieval}},
  language     = {{eng}},
  pages        = {{338--347}},
  publisher    = {{IEEE Computer Society}},
  title        = {{Active Exploration for Sparse Visual Localization}},
  url          = {{https://openaccess.thecvf.com/content/CVPR2026F/papers/Lidholm_Active_Exploration_for_Sparse_Visual_Localization_CVPRF_2026_paper.pdf}},
  year         = {{2026}},
}