b-FLAIR-test: Building Change Detection Evaluation Dataset

Dataset Description

b-FLAIR-test is an evaluation dataset for building change detection, containing 1,730 annotated image pairs with binary building change masks. This dataset is designed for in-domain evaluation of methods trained on b-FLAIR in particular and provides a rigorous benchmark for bi-temporal building change detection in general.

Project page: https://xavibou.github.io/CDviaWTS/

Dataset Format

• Number of pairs: 1,730 image pairs
• Image format: 5-band images (Red, Green, Blue, Infrared, Elevation), 512×512 pixels
• Resolution: 0.2 meters per pixel
• Annotation: Binary building change masks
• Geographic coverage: 9 different French administrative departments
• Change types: New building constructions or no change (~30% of pairs show no change)

Key Features

• Images processed and formatted following the FLAIR dataset [1] procedure from BD ORTHO imagery [2]
• Expert-annotated pairs verified by independent assessors
• Focuses exclusively on building construction (no building destruction cases included)
• Compatible with models trained on FLAIR [1] or b-FLAIR datasets

References

[1] Garioud et al. (2023). FLAIR: a country-scale land cover semantic segmentation dataset from multi-source optical imagery. In NeurIPS
[2] IGN - Institut national de l’information géographique et forestière. (2025). BD ORTHO®: L’image géographique du territoire national, la France vue du ciel.

Citation

If you use this dataset, please cite the following publication:

@article{bou2026remote,
  title={Remote Sensing Change Detection via Weak Temporal Supervision},
  author={Bou, Xavier and Vincent, Elliot and Facciolo, Gabriele and Grompone von Gioi, Rafael and Morel, Jean-Michel and Ehret, Thibaud},
  journal={arXiv preprint arXiv:2601.02126},
  year={2026}
}