ViDSOD-100
收藏资源简介:
ViDSOD-100是由香港科技大学(广州)等机构创建的RGB-D视频显著对象检测数据集,包含100个视频共9,362帧,覆盖多种自然场景。数据集中的每一帧都经过人工标注,提供了高质量的显著性标注。该数据集的创建旨在解决动态RGB-D视频中显著对象检测的挑战,特别是显著性随时间变化,即显著性转移的问题。ViDSOD-100数据集的应用领域包括图像理解、动作识别等,为研究社区提供了一个新的学习和研究平台。
ViDSOD-100 is an RGB-D video salient object detection dataset created by institutions including The Hong Kong University of Science and Technology (Guangzhou). It consists of 100 videos totaling 9,362 frames, covering a wide range of natural scenes. Every frame in the dataset is manually annotated with high-quality saliency annotations. This dataset is developed to address the challenges in salient object detection for dynamic RGB-D videos, particularly the issue of temporal variation of saliency, namely saliency shift. ViDSOD-100 has applications in fields such as image understanding and action recognition, providing a new learning and research platform for the research community.




