Additional BDD100K Context Annotations
收藏资源简介:
This dataset provides additional context annotations and extends the BDD100K dataset [1]. Therefore, the approximately 80,000 images with annotation for 2D object detection in the BDD100K dataset were annotated with additional context attributes. The application possibilities of the dataset are diverse. It could be used for model training or evaluating the model performance in different contexts combination, as we did. The annotated contexts per image contain 11 attributes (time_of_day, sky, illumination, precipitation, infrastructure, road, tunnel, construction_site, clear_windshield, light_exposure, and reflections). The underlying annotation guideline with further details is also available for download.
本数据集提供额外的上下文标注,并对BDD100K数据集[1]进行了扩展。具体而言,原BDD100K数据集中用于二维目标检测(2D object detection)的约8万张带标注图像,新增了上下文属性标注。本数据集的应用场景广泛多元,既可用于模型训练,也可如本研究所开展的那样,在不同上下文组合下评估模型性能。每张图像的标注上下文涵盖11项属性:时段(time_of_day)、天空状况(sky)、光照条件(illumination)、降水情况(precipitation)、基础设施(infrastructure)、道路状态(road)、隧道场景(tunnel)、施工场地(construction_site)、挡风玻璃清洁度(clear_windshield)、光照暴露度(light_exposure)以及反射情况(reflections)。附带详细细节的标注指南同样可供下载。




