Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

DanceTrack

DanceTrack 是一个大规模多目标跟踪(MOT)数据集,专注于统一外观与多样化运动场景下的人体跟踪任务。数据集由群舞视频构成,舞者穿着高度相似甚至相同的服装,伴随复杂多变的肢体动作、频繁遮挡与交叉。该数据集发表于 CVPR 2022。

设计目标:促使 MOT 算法减少对视觉外观判别的依赖,更多利用运动分析完成目标关联。

关键特性

  • 统一外观:人物穿着高度相似甚至相同,视觉特征难以区分。
  • 多样化运动:大幅度的运动范围与复杂姿态变化,对运动建模提出更高要求。

支持的任务

  • 多目标跟踪(MOT)
  • 目标检测(作为跟踪的基础)
  • 行人重识别(在相似外观条件下的评估)

目录结构

{DanceTrack ROOT}
└── dancetrack/
    ├── train/          # 40 个序列,标注公开
    │   └── dancetrack0001/
    │       ├── img1/   # 图像序列(JPEG)
    │       └── gt/gt.txt
    ├── val/            # 25 个序列,标注公开
    └── test/           # 35 个序列,无标注(需提交评测)

标注格式(gt.txt)

每行:<frame_id>, <track_id>, <box_left>, <box_top>, <box_right>, <box_bottom>, 1, 1, 1

  • 坐标为绝对像素值,边界框为左上/右下角。
  • frame_id 从 1 开始;后三个 1 为固定占位符。
  • 标注规范:部分遮挡时标注完整人体框,完全遮挡时不标注。

数据划分

划分 视频数 标注状态
Train 40 公开
Validation 25 公开
Test 35 不公开

测试集需通过 CodaBench 提交评测。

引用

@inproceedings{sun2022dance,
  title={DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion},
  author={Sun, Peize and Cao, Jinkun and Jiang, Yi and Yuan, Zehuan and Bai, Song and Kitani, Kris and Luo, Ping},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2022}
}

License

本数据集为非商业研究用途镜像,原始协议如下:

  • 标注(annotations):采用 Creative Commons Attribution 4.0 (CC BY 4.0) 许可。
  • 数据本体(videos / images):仅限非商业研究用途。
  • 所有视频与图像均来自互联网,其版权不属于香港大学、卡内基梅隆大学或字节跳动;上述机构不对这些视频与图像的内容或含义负责。

原始发布方声明见 https://dancetrack.github.io/,权威协议以原始页面为准。本镜像不授予超出原始协议的额外权利。

Downloads last month
74