CrowdTrack
收藏资源简介:
CrowdTrack 是一个大规模的多行人跟踪数据集,包含33个视频序列,约4万张图像帧,以及超过70万个行人标注。该数据集旨在解决现实世界跟踪场景中的关键问题,涵盖了各种复杂的场景,如人群密集、遮挡和模糊等,为算法研究提供了丰富的训练信号,以提高模型在现实世界复杂性下的鲁棒性。数据集包含了多样化的相机设置,包括移动和固定镜头拍摄,所有数据均来自无约束的日常生活环境,确保了行人行为的自然性,从而增强了数据集在实际应用中的相关性,如机器人技术和自动驾驶。
CrowdTrack is a large-scale multi-person tracking dataset containing 33 video sequences, approximately 40,000 image frames, and over 700,000 pedestrian annotations. This dataset aims to address key challenges in real-world tracking scenarios, covering various complex situations such as dense crowds, occlusions, and motion blurs. It provides abundant training signals for algorithm research to improve the robustness of models against real-world complexities. The dataset includes diverse camera configurations, with both mobile and fixed-lens captures. All data are collected from unconstrained daily-life environments, ensuring the naturalness of pedestrian behaviors, thereby enhancing the practical relevance of the dataset for applications such as robotics and autonomous driving.
CrowdTrack数据集概述
数据集基本信息
- 官方仓库:CrowdTrack数据集是论文《CrowdTrack: A Benchmark for Difficult Multiple Pedestrian Tracking in Real Scenarios》的官方数据集。
- 数据集地址:Baidu Cloud
密码:crtr
数据集内容
- 场景分布:包含场景分布的可视化图表(见assets/changjing.png)。
- 统计特征:包含数据集的统计特征可视化图表(见assets/tongji.png)。
相关论文
- 论文链接:预印本论文可在arXiv获取:arXiv:XXXX.XXXXX
排行榜
- 排行榜信息:包含排行榜的可视化图表(见assets/leaderboard.png)。




