MOSE
收藏资源简介:
MOSE数据集是由南洋理工大学等机构创建,专注于复杂场景下的视频对象分割。该数据集包含2,149个视频片段,总计5,200个对象,覆盖36个类别,拥有431,725个高质量对象分割掩码。MOSE的显著特点是包含大量拥挤和遮挡对象,目标对象常被其他对象遮挡或在某些帧中消失。数据集的创建过程涉及从现有数据集继承视频和从现实世界场景新拍摄视频,确保包含多种遮挡场景、运动场景和对象消失-重现场景。MOSE数据集旨在推动更全面和鲁棒的视频对象分割算法的发展,特别是在处理复杂环境中的对象跟踪和分割问题。
The MOSE dataset was developed by institutions including Nanyang Technological University, focusing on video object segmentation in complex real-world scenarios. It contains 2,149 video clips, with a total of 5,200 objects spanning 36 categories, and includes 431,725 high-quality object segmentation masks. A prominent characteristic of the MOSE dataset is its abundant coverage of crowded and occluded objects, where target objects are often blocked by other entities or vanish in certain frames. The dataset construction process involves sourcing videos both by inheriting from existing datasets and filming new footage in real-world scenes, ensuring that diverse scenarios such as occlusions, object motions, and object disappearance-reappearance events are fully covered. The MOSE dataset aims to facilitate the advancement of more comprehensive and robust video object segmentation algorithms, particularly for tackling object tracking and segmentation challenges in complex environments.

- 1MOSE: A New Dataset for Video Object Segmentation in Complex Scenes南洋理工大学 · 2023年



