MOSEv2
收藏资源简介:
MOSEv2数据集是一个用于视频目标分割的更具有挑战性的数据集,旨在推动视频目标分割方法在现实世界环境中的进一步发展。该数据集包含5024个视频和超过701,976个高质量的面具,涉及10,074个物体和200个类别。与它的前身MOSEv1相比,MOSEv2引入了更大的场景复杂性,包括更频繁的物体消失和重新出现,严重的遮挡和拥挤,更小的物体,以及一系列新的挑战,如恶劣天气、低光场景、多镜头序列、伪装物体、非物理目标(如阴影、反射)和需要外部知识的场景等。这些多样化和细粒度的挑战使MOSEv2成为评估和推进开放世界复杂场景中视频目标分割鲁棒性的理想基准。
MOSEv2 is a more challenging dataset for video object segmentation, designed to advance the development of video object segmentation methods in real-world environments. This dataset includes 5,024 videos and over 701,976 high-quality masks, encompassing 10,074 objects and 200 categories. Compared with its predecessor MOSEv1, MOSEv2 features greater scene complexity, including more frequent object disappearance and reappearance, severe occlusion and crowding, smaller objects, as well as a series of new challenges such as adverse weather, low-light scenarios, multi-shot sequences, camouflaged objects, non-physical targets (e.g., shadows, reflections), and scenarios requiring external knowledge, etc. These diverse and fine-grained challenges render MOSEv2 an ideal benchmark for evaluating and advancing the robustness of video object segmentation models in complex open-world scenarios.
MOSEv2数据集概述
基本信息
- 许可证: CC BY-SA 4.0
- 任务类别: 目标检测
- 标签: 视频对象分割、复杂场景、计算机视觉、分割、视频分析、基准测试
- 数据规模: 1K<n<10K
- 语言: 英语
- 名称: MOSEv2
- 论文: arXiv:2508.05630
数据集简介
MOSEv2是一个全面的视频对象分割数据集,旨在推动真实世界条件下的VOS方法发展。包含:
- 5,024个视频
- 701,976个高质量掩码
- 10,074个对象
- 200个类别
数据集特点
- 场景复杂性:
- 更频繁的对象消失和重现
- 严重的遮挡和拥挤
- 更小的对象
- 恶劣天气条件(雨、雪、雾)
- 低光场景(夜间、水下)
- 多镜头序列
- 伪装对象
- 非物理目标(阴影、反射)
- 需要外部知识的场景
基准测试结果
- VOS方法性能下降:
- SAM2: 76.4% (MOSEv1) → 50.9% (MOSEv2)
- 视频对象跟踪方法也观察到类似下降
数据集结构
<train/valid.tar.gz> │ ├── Annotations │ │ │ ├── <video_name_1> │ │ ├── 00000.png │ │ ├── 00001.png │ │ └── ... │ │ │ ├── <video_name_2> │ │ ├── 00000.png │ │ ├── 00001.png │ │ └── ... │ │ │ ├── <video_name_...> │ └── JPEGImages │ ├── <video_name_1> │ ├── 00000.jpg │ ├── 00001.jpg │ └── ... │ ├── <video_name_2> │ ├── 00000.jpg │ ├── 00001.jpg │ └── ... │ └── <video_name_...>
引用
bibtex @article{MOSEv2, title={{MOSEv2}: A More Challenging Dataset for Video Object Segmentation in Complex Scenes}, author={Ding, Henghui and Ying, Kaining and Liu, Chang and He, Shuting and Jiang, Xudong and Jiang, Yu-Gang and Torr, Philip HS and Bai, Song}, journal={arXiv preprint arXiv:2508.05630}, year={2025} }
许可证
- 类型: CC BY-NC-SA 4.0
- 限制: 仅限非商业研究用途

- 1MOSEv2: A More Challenging Dataset for Video Object Segmentation in Complex Scenes复旦大学,字节跳动公司,上海财经大学,南洋理工大学,牛津大学 · 2025年



