Vojtěch Pánek

I am a PhD student at CIIRC CTU in Prague, under the supervision of Torsten Sattler. Currently, my main focus is on visual localization.

I've studied my Bachelor's and Master's degrees at FEE, CTU in Prague, dealing with a generation of NDT maps from CAD floor plans and visual robot localization using an omnidirectional camera.

Email  /  CV  /  Google Scholar  /  Github

profile photo
Research

At the moment, I focus mainly on exploring the use of non-standard representations for the visual localization task.

clean-usnob Visual Localization using Imperfect 3D Models from the Internet
Vojtech Panek, Zuzana Kukelova, Torsten Sattler
The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023
arXiv  /  benchmark  /  data  /  bibtex

Exploring the use of 3D models from the Internet for visual localization and studies studies how the imperfections of these models affect localization accuracy. Accompanied by the benchmark.

clean-usnob MeshLoc: Mesh-Based Visual Localization
Vojtech Panek, Zuzana Kukelova, Torsten Sattler
European Conference on Computer Vision (ECCV), 2022
arXiv  /  code  /  data  /  bibtex

Visual localization pipeline storing just 3D mesh as the representation of the environment. Rendering virtual views and using modern local features on-the-fly. Suprisingly can work somehow even on meshes without any color information.

clean-usnob Visual Localization with Environment Outline Prior
Vojtech Panek, supervised by Torsten Sattler
International Student Conference on Electrical Engineering (POSTER), 2022
paper  /  poster

Two-stage visual localization pipeline with environment outlines (floor plans / buildings from OSM) as environment representation. Using retrieval on set of local outlines for coarse pose estimate and ICP for pose refinement.

Teaching

I am a teaching assistant for a few courses at the Faculty of Electrical Engineering, CTU in Prague:

Check CourseWare for up-to-date information on the courses.


The original design of the website comes from Jon Barron. See his GitHub repository for the source code.