Annotations for training dataset of images of birds in flight labelled with upstroke and downstroke

Annotations for a training dataset for a model that can classify between images of birds flying in upstroke and birds flying in downstroke. The corresponding image dataset is formed from 8,699 images of birds in flight, manually selected from the North American Birds dataset (NABirds) and the 2019 iNaturalist dataset. We manually annotated each bird within the selected images from these datasets (some images had more than one bird) as being either in upstroke or downstroke to produce a new labelled dataset. There are 5,754 downstrokes and 4,421 upstrokes. As we are unable to distribute the images these annotations correspond to, please find them through the links in Related works. The filenames of the annotation files will correspond to the filenames of the original images. Labelling was achieved using LabelImg and model training was done through the Python detecto module. To use these labels through detecto, follow the README at the linked GitHub.本数据集标注对应一款可对振翅上行与振翅下行阶段的鸟类飞行图像进行分类的模型训练数据集。 对应的图像数据集共包含8699张飞行鸟类图像，均从北美鸟类数据集（North American Birds, NABirds）与2019年iNaturalist数据集（iNaturalist 2019）中手动筛选得到。我们对筛选出的每张图像中的每只鸟类（部分图像包含多只鸟类）进行手动标注，标记其处于上挥冲程（upstroke）或下扑冲程（downstroke），从而构建得到全新的带标注数据集。本数据集共包含5754个下扑冲程样本与4421个上挥冲程样本。 因无法随标注一同发布对应的原始图像，请通过相关工作（Related works）中提供的链接获取图像。标注文件的文件名将与原始图像的文件名一一对应。 本次标注工作采用LabelImg工具完成，模型训练则通过Python的detecto模块实现。若需通过detecto模块使用本数据集标注，请参照对应GitHub仓库中的README文档。