Practical solutions to the relative pose of three calibrated cameras

This repo contains code for paper "Practical solutions to the relative pose of three calibrated cameras" (CVPR 2025 - paper link)

Installation

Create an environment with pytorch and packaged from requirements.txt.

Install PoseLib fork with implemented estimators in branch threeview into the environment:

git clone https://github.com/kocurvik/PoseLib
git cd PoseLib
git checkout threeview
pip install .

Before running the python scripts make sure that the repo is in your python path (e.g. export PYTHONPATH=/path/to/repo/threeview)

Data preparation

Triplets can be made using script dataset/prepare_im.py from a dataset with a Colmap reconstruction available.

You can also download the triplets already extracted for Phototourism and Aachen.

Evaluation

To perform the evaluation on real data run for each scene:

python eval.py -nw 64 triplets-features_superpoint_noresize_2048-LG /path/to/scene_folder_with_triplets

Citation

@inproceedings{tzamos2025practical,
  title={Practical solutions to the relative pose of three calibrated cameras},
  author={Tzamos, Charalambos and Kocur, Viktor and Ding, Yaqing and Barath, Daniel and Haladov{\'a}, Zuzana Berger and Sattler, Torsten and Kukelova, Zuzana},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={21913--21923},
  year={2025}
}