b-FLAIR-spot: bi-temporal extension of FLAIR in SPOT-6/7 modality

Dataset Description

b-FLAIR-spot is a temporal extension of the FLAIR dataset [1], mirroring b-FLAIR in SPOT-6/7 modality, focused on land cover classification in France. The dataset provides bi-temporal satellite image pairs with single-temporal semantic annotations.

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

Dataset Summary

• Task: Semantic change detection via weak temporal supervision
• Coverage: France
• Resolution: 1.6 m/px
• Patch Size: 64×64 pixels
• Bands: Red, Green, Blue
• Total Training Pairs: 61,712
• Total Test Pairs: 16,050
• Validation split: Corresponds to folders D004, D014, D029, D031, D058, D066, D067, D077 in train

Dataset Creation

Source Data

The dataset is a mirror of b-FLAIR in SPOT-6/7 modality, further extending the original FLAIR dataset [1]. New images were downloaded from IGN's ORTHO-SAT database [2], at a original resolution of 1.5 meters per pixel, and were resampled at the resolution of 1.6 meters per pixel.

Annotations

Original FLAIR single-temporal semantic masks are provided for each pair. They classify each pixel of t1 images in one of 19 semantic land cover classes and were resampled at the resolution of 1.6 meters per pixel. Please refer to [1] for more information on these annotations.

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). ORTHO-SAT®: Les ortho-images issues de prises de vues satellitaires

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}
}