OVIS
收藏资源简介:
OVIS数据集是由华中科技大学和阿里巴巴集团联合创建的大规模视频实例分割数据集,专注于在遮挡场景中同时检测、分割和跟踪实例。该数据集包含296,000个高质量实例掩码,涵盖25个语义类别,其中大部分对象存在遮挡情况。OVIS数据集的创建旨在探索视频理解系统在处理实际世界中广泛存在的遮挡问题时的能力,特别是在视频实例分割任务中。数据集的访问地址为http://songbai.site/ovis,适用于评估和开发处理严重对象遮挡的视频实例分割模型。
The OVIS dataset is a large-scale video instance segmentation dataset jointly created by Huazhong University of Science and Technology and Alibaba Group, which focuses on simultaneously detecting, segmenting and tracking instances in occluded scenarios. It contains 296,000 high-quality instance masks covering 25 semantic categories, with most of the objects experiencing occlusion. The OVIS dataset was developed to investigate the performance of video understanding systems in addressing the ubiquitous occlusion issues in real-world scenarios, with a particular focus on the video instance segmentation task. The dataset is accessible at http://songbai.site/ovis, and is suitable for evaluating and developing video instance segmentation models that handle severe object occlusion.




