Datasets:

Modalities:
Image
ArXiv:
Libraries:
Datasets
License:
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.

MOT20

MOT20 是 MOTChallenge 多目标跟踪基准的 2020 年版,专门面向极高人群密度场景,用于推进遮挡条件下的行人跟踪研究。

数据构成

  • 8 个视频序列(4 个训练 + 4 个测试):MOT20-01 ~ MOT20-08。
  • 场景为极拥挤的公共场所(广场、车站等),单帧目标数可达数百。
  • 本仓库包含:
    文件 说明
    MOT20.zip 图像序列
    MOT20Labels.zip 训练集标注

目录结构

{MOT20 ROOT}
├── train/
│   └── MOT20-01/
│       ├── img1/            # 图像序列
│       ├── det/det.txt      # 公开检测结果
│       └── gt/gt.txt        # 标注
└── test/
    └── MOT20-06/
        ├── img1/
        └── det/det.txt

标注格式(gt.txt)

每行 10 个字段: <frame>, <id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, <class>, <visibility>, <ignore>

  • frame、id 从 1 开始;conf = 0 表示该条目不参与评测。
  • 2D 跟踪任务中世界坐标 x, y, z 可填 -1。

评测

官方评测请通过 MOTChallenge 平台或 CodaBench 提交,工具推荐 TrackEval。

引用

@article{dendorfer2020mot20,
  title={MOT20: A Benchmark for Multi Object Tracking in Crowded Scenes},
  author={Dendorfer, Patrick and Rezatofighi, Hamid and Milan, Anton and Shi, Javen and Cremers, Daniel and Reid, Ian and Roth, Stefan and Schindler, Konrad and Leal-Taix{\'e}, Laura},
  journal={arXiv preprint arXiv:2003.09003},
  year={2020}
}

License

CC BY-NC-SA 3.0

MOTChallenge 数据集的官方声明为 Creative Commons Attribution-NonCommercial-ShareAlike 3.0 License:

  • 须按作者指定的方式署名;
  • 不得用于商业目的;
  • 若修改、转换或基于本数据创作,衍生作品须采用相同许可分发。

协议全文:https://creativecommons.org/licenses/by-nc-sa/3.0/

使用与隐私

MOT20 包含真实公共场所的行人影像。使用者在训练与部署模型时需考虑隐私、监控与公平性影响。已知局限:测试集标注不公开;场景不覆盖所有国家、相机类型、光照与人群构成;模型在真实监控或安全关键场景部署前应谨慎评估。

权威访问条款请以 https://motchallenge.net/data/MOT20/ 为准,本镜像不授予超出原始协议的额外权利。

Downloads last month
90

Paper for mrhwl123/MOT20