FLAIR-HUB
收藏资源简介:
FLAIR-HUB是法国国家地理与森林信息研究所(IGN)推出的一个大规模多模态土地覆盖数据集,拥有超高分辨率(20厘米)的标注,覆盖了法国2528平方公里的区域。该数据集结合了六种对齐的模态,包括航空影像、Sentinel-1/2时间序列、SPOT影像、地形数据以及历史航空影像。FLAIR-HUB包含超过630亿个像素的手工标注,涵盖了19个土地覆盖类别和23种作物类型。数据集的空间分辨率对于土地覆盖分析至关重要,可以精确测量地表和边界,捕捉到诸如房屋、树木或道路等小规模特征。FLAIR-HUB旨在支持监督学习和多模态预训练,并促进大规模语义分割方法的发展。该数据集可用于多模态自我监督方法或数据融合方法的研究,并且将随着新对齐模态或新标注的添加而不断发展。
FLAIR-HUB is a large-scale multimodal land cover dataset launched by the French National Institute of Geographic and Forestry Information (IGN). It features ultra-high-resolution (20 cm) annotations and covers an area of 2528 square kilometers across France. This dataset integrates six aligned modalities, including aerial imagery, Sentinel-1/2 time series, SPOT imagery, topographic data, and historical aerial imagery. FLAIR-HUB contains over 63 billion manually annotated pixels, covering 19 land cover categories and 23 crop types. The spatial resolution of the dataset is critical for land cover analysis, enabling precise measurement of land surfaces and boundaries, as well as capturing small-scale features such as buildings, trees, and roads. FLAIR-HUB aims to support supervised learning and multimodal pre-training, and advance the development of large-scale semantic segmentation methods. It can be utilized for research on multimodal self-supervised methods or data fusion approaches, and will continue to evolve with the addition of new aligned modalities or new annotations.

- 1FLAIR-HUB: Large-scale Multimodal Dataset for Land Cover and Crop Mapping法国国家地理与森林信息研究所(IGN) · 2025年



