WXSOD
收藏资源简介:
WXSOD是一个用于鲁棒性显著物体检测的数据集,旨在帮助研究在恶劣天气条件下的显著物体检测问题。数据集包含14,945张RGB图像,覆盖超过5,000个场景,包含各种天气噪声,并分为合成训练集、合成测试集和真实测试集。该数据集的设计旨在减少标签成本并确保场景多样性,同时考虑了由天气噪声引起的域差异问题。WXSOD是首个针对恶劣天气条件下RGB显著物体检测的基准数据集,它涵盖了不同类别的显著物体、不同尺寸和数量的显著物体,以及多种天气噪声。数据集的创建过程包括从现有数据集中选择户外场景,并利用图像风格转换库模拟各种天气噪声。此外,WXSOD还包括了从互联网上收集的554张真实场景图像,以增强数据集的真实性。WXSOD的应用领域是显著物体检测,旨在解决在恶劣天气条件下显著物体检测的问题。
WXSOD is a dataset for robust salient object detection, designed to facilitate research on salient object detection under adverse weather conditions. It contains 14,945 RGB images covering over 5,000 scenarios with various weather noises, and is split into three subsets: synthetic training set, synthetic test set, and real test set. The dataset is engineered to reduce labeling costs and ensure scenario diversity, while addressing the domain shift issue induced by weather noises. As the first benchmark dataset for RGB salient object detection under adverse weather conditions, WXSOD covers salient objects of different categories, varying sizes and quantities, as well as multiple types of weather noises. The dataset construction process includes selecting outdoor scenarios from existing datasets and simulating various weather noises via image style transfer libraries. Additionally, WXSOD incorporates 554 real-scenario images collected from the Internet to further improve its realism. The target application of WXSOD is salient object detection, specifically addressing the challenges of salient object detection under adverse weather conditions.




