ESA-PhiLab-Edge/OEOBench-WorldFloods

WorldFloods-PhiSat-2数据集是WorldFloods v2数据集的一个改编版本，旨在模拟Φsat-2类似的数据集，用于训练该任务的基础模型。数据集使用OrbitalAI Challenge中的代码模拟Φsat-2图像，并遵循PhiSatNet存储库的数据规范格式进行存储。由于WorldFloods v2数据集包含完整的洪水场景而非小斑块，因此从模拟图像中创建了尺寸为512x512像素的非重叠训练和验证斑块，并且只存储云覆盖率小于20%的斑块，以便实时训练论文中描述的洪水分割模型。对于测试样本，保留了完整场景，以便更好地评估洪水检测模型。数据集的总大小为389 GB。

The WorldFloods Φsat-2 dataset is an adaptation of the WorldFloods v2 dataset created with the purpose of simulating a Φsat-2-like dataset for training Foundation Models for this mission. Φsat-2 images were simulated using the code in the OrbitalAI Challenge and stored following the PhiSatNet Data Specification Format. Since the WorldFloods v2 dataset contains full flood scenes rather than small patches, non-overlapping Train and Validation patches of size 512x512 pixels were created from the simulated images, and only patches with less than 20% cloud cover were stored to train the flood segmentation models described in the paper Global flood extent segmentation in optical satellite images. For Test samples, however, the full scenes were kept to allow for better evaluation of flood detection models. The total size of the dataset is 389 GB.