CoInfra
收藏资源简介:
CoInfra 是一个大规模的协作基础设施感知系统和数据集,旨在推进在真实世界和恶劣天气条件下的鲁棒多智能体感知。该数据集包括来自8个基础设施节点的195k LiDAR帧和390k摄像头图像,这些节点在全球范围内时间同步和空间校准。数据集收集了不同天气场景下的数据,包括晴天、雨天、冻雨和大雪。此外,提供了5个物体类别(即汽车、公共汽车、卡车、人和自行车)的全面3D边界框注释,以及高清晰度地图,以便进行上下文理解。该数据集旨在解决在复杂城市环境中恶劣天气条件下准确感知环境和所有参与交通代理的关键挑战。
CoInfra is a large-scale collaborative infrastructure-aware system and dataset, which aims to advance robust multi-agent perception in real-world and adverse weather conditions. This dataset contains 195k LiDAR frames and 390k camera images from 8 infrastructure nodes that are temporally synchronized and spatially calibrated globally. It collects data under diverse weather scenarios including sunny days, rainy days, freezing rain and heavy snow. In addition, comprehensive 3D bounding box annotations for 5 object categories (i.e., car, bus, truck, pedestrian and bicycle) as well as high-definition maps are provided to facilitate contextual understanding. This dataset is designed to address the critical challenge of accurately perceiving the environment and all involved traffic agents in complex urban environments under adverse weather conditions.
CoInfra数据集概述
数据集简介
- 名称:CoInfra
- 类型:大规模协同基础设施感知系统与数据集
- 特点:专为恶劣天气条件(雪、雨、冻雨)下的多智能体感知研究设计
- 规模:
- 14个同步基础设施节点
- 每个节点配备1个LiDAR和2个摄像头
- 覆盖共享区域的全场景感知
数据集内容
- 数据量:
- 约195k LiDAR帧
- 约390k相机图像
- 220k标注的3D边界框(5类:汽车、公交车、卡车、自行车、行人)
- 天气场景:晴天、雨天、雪天、冻雨
- 附加数据:
- HD地图
- 相机-LiDAR标定数据
- 全局对齐数据
下载信息
- 示例数据集:Mini Example Dataset
- 完整数据集获取方式:
- 邮箱联系:MVS-Lab@uwaterloo.ca
- HuggingFace下载:Download
应用场景
- 恶劣天气下的多节点3D目标检测与跟踪
- 延迟感知的实时感知同步
- 交互密集环岛场景的运动预测
基准实验
| 融合策略 | HD地图 | mAP(所有类别) |
|---|---|---|
| 早期融合 | ✘ | 0.984 |
| 早期融合 | ✔️ | 0.986 |
| 后期融合 | ✘ | 0.931 |
| 后期融合 | ✔️ | 0.952 |
引用
bibtex @article{ning2025coinfra, title={CoInfra: A Large-Scale Cooperative Infrastructure Perception System and Dataset in Adverse Weather}, author={Ning, Minghao and Yang, Yufeng and Shu, Keqi and Huang, Shucheng and Zhong, Jiaming and Salehi, Maryam and Rahmani, Mahdi and Lu, Yukun and Sun, Chen and Saleh, Aladdin and Hashemi, Ehsan and Khajepour, Amir}, journal={arXiv preprint arXiv:2507.02245}, url={https://arxiv.org/abs/2507.02245}, year={2025} }
许可证
- 类型:CC BY-NC 4.0 License




