Flair2 Challenge

CNES and ENSG (Ecole Nationale des Sciences Géographiques) launched in 2023 the Flair2 challenge, an initiative aimed to enhance the Earth Observation community's ability to map and monitor the impact of human activities on the biosphere. It focused on helping the European community understand France's state-of-the-art approaches to addressing soil artificialisation challenges.

For this initiative, CNES offered a land cover mapping pipeline using deep learning and high-quality ground truth data. This mapping was accessible to participants in a 4 months long challenge launched by the National Institute for Geographic and Forest Information (IGN/ENSG), promoting the use of Copernicus satellite data and high-resolution aerial photography. CNES participated in its success by opening up this challenge on a European scale and contributing to the setting up and payment of the award.

More than 100 persons from various European countries participated to the challenge.

The main achievements are:

• The Flair2 challenge took place from May 25th to September 25th 2023. CNES, in collaboration with IGN and ENSG, developed a comprehensive communication program to:
  • Clearly and visually present the challenges of land use.
  • Showcase the data used by the French National Mapping Institute to monitor land-take.
  • Explain the technical challenges of advanced, large-scale deep learning.
  • Address the difficulty of integrating these aspects.
  • Identify European institutions for communication.
  • Guide participation and access to the necessary code and data for the challenge.
  • Produce scientific papers to share these insights with the scientific community.

Links & Publications:
Due to the community’s interest in the topic and the originality of the Datafusion approach, 4 papers were written:

• Anatol Garioud, Nicolas Gonthier, Loic Landrieu, Apolline De Wit, Marion Valette, Marc Poupée, Sébastien Giordano and Boris Wattrelos.
  FLAIR: a Country-Scale Land Cover Semantic Segmentation Dataset From Multi-Source Optical Imagery. (2023).
  In proceedings of Advances in Neural Information Processing Systems (NeurIPS) 2023.
  DOI: https://doi.org/10.48550/arXiv.2310.13336
• Anatol Garioud, Apolline De Wit, Marc Poupée, Marion Valette, Sébastien Giordano, Boris Wattrelos.
  FLAIR #2: textural and temporal information for semantic segmentation from multi-source optical imagery. (2023)
  arXiv cs.CV 2305.14467
  DOI : https://doi.org/10.13140/RG.2.2.30938.93128/1
• Jakub Straka, Ivan Gruber.
  Modernized Training of U-Net for Aerial Semantic Segmentation. (2024)
  Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops, 776–784.
  URL : https://openaccess.thecvf.com/content/WACV2024W/CV4EO/html/Straka_Modernized_Training_of_U-Net_for_Aerial_Semantic_Segmentation_WACVW_2024_paper.html
• Guillaume Astruc, Nicolas Gonthier, Clement Mallet, Loic Landrieu.
  OmniSat: Self-Supervised Modality Fusion for Earth Observation. (2024)
  arXiv cs.CV, proofreading for an A-rank conference.
  URL : https://arxiv.org/abs/2404.08351