cloudwalk-research/sp-swimming-pools

São Paulo游泳池检测数据集是一个基于航空影像的目标检测数据集，专注于巴西圣保罗市的游泳池。数据集从GeoSampa正射影像（约0.4米/像素）中切割成512×512的图块，并通过两种方式进行标注：一种是使用Roboflow检测工作流（general-segmentation-api-3，置信度0.25），另一种是使用在这些Roboflow标签上训练的pool_v4 YOLOv11检测器。README明确指出所有标签均为模型生成，未经人工验证，因此被视为“银标准”。数据集包含三个分割（train、val、weak），分别用于训练、验证和弱监督学习。此外，README还详细介绍了数据集的模式、标签来源、AOI选择方法、引用信息、数据来源和许可证。

The São Paulo Swimming Pool Detection dataset is an aerial-imagery object-detection dataset focused on swimming pools across the city of São Paulo, Brazil. It is built from GeoSampa orthophoto imagery (approximately 0.4 m/px) chipped into 512 × 512 tiles and labeled in two ways: by a hosted Roboflow detection workflow (general-segmentation-api-3 at confidence 0.25) and by the pool_v4 YOLOv11 detector trained on those Roboflow labels. The README highlights that all labels are model-generated without manual ground truth, classifying them as silver-standard. The dataset includes three splits (train, val, weak) for training, validation, and weak supervision. The README also provides detailed information on the datasets schema, label provenance, AOI selection, citation, source, and license.