[WACV 2026] Beyond Paired Data: Self-Supervised UAV Geo-Localization from Reference Imagery Alone

Official repository of the paper:
"Beyond Paired Data: Self-Supervised UAV Geo-Localization from Reference Imagery Alone"
Tristan Amadei, Enric Meinhardt-Llopis, Benedicte Bascle, Corentin Abgrall, Gabriele Facciolo
IEEE/CVF Winter Conference on Applications of Computer Vision 2026

This repository contains the code and instructions to reproduce the experiments of CAEVL, a data-efficient UAV geolocalization method trained without paired UAV–satellite images. Our method leverages edge-based representations and non-contrastive learning to achieve competitive performance while being lightweight and generalizable.

---

📦 ViLD Dataset

We are preparing the ViLD dataset for release - stay tuned!

---

Citation

@article{amadei2026caevl,
  title={Beyond Paired Data: Self-Supervised UAV Geo-Localization from Reference Imagery Alone},
  author={Amadei, Tristan and Meinhardt-Llopis, Enric and Bascle, Benedicte and Abgrall, Corentin and Facciolo, Gabriele},
  journal={2026 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year={2026},
  organization={IEEE}
}

Acknowledge

We thank the contributors of the open source codes including Dinov2 (https://github.com/facebookresearch/dinov2) and (https://github.com/gmberton/VPR-methods-evaluation).