OpenEarthMap Land Cover Mapping Few-Shot Challenge

该数据集是第三届有限标注数据图像与视频理解研讨会（L3D-IVU）的一部分，该研讨会与2024年CVPR会议同期举行。数据集包含高分辨率的土地利用覆盖映射图像，这些图像用于一项挑战，旨在有限的标注数据情况下发现新的土地利用覆盖类别。具体来说，该数据集包括一个基础训练集，含有258个图块，以及一个针对新类别的5次射击支持集。在规模上，基础训练集有258个图块，而支持集则有20个图块。该数据集的任务是土地利用覆盖的分类与分割。

This dataset is part of the 3rd Workshop on Limited Labeled Data for Image and Video Understanding (L3D-IVU), which is co-located with the 2024 Conference on Computer Vision and Pattern Recognition (CVPR). This dataset contains high-resolution land use/land cover (LULC) mapping images, which are used for a challenge that aims to discover novel land use/land cover categories under the setting of limited labeled data. Specifically, this dataset includes a base training set with 258 image tiles, and a 5-shot support set for novel categories. In terms of scale, the base training set consists of 258 tiles, while the support set contains 20 tiles. The task of this dataset is land use/land cover classification and segmentation.