V2X-Seq: A Large-Scale Sequential Dataset for Vehicle-Infrastructure Cooperative Perception and Forecasting

• DOI: 10.48550/arxiv.2305.05938
• Publication Date: 2023-01-01
• Authors: Haibao Yu; Wenxian Yang; Hongzhi Ruan; Z. Yang; Yingjuan Tang; Xu Gao; Hao Xin; Yifeng Shi; Yifeng Pan; Ning Sun; Juan Song; Jirui Yuan; Ping Luo; Zaiqing Nie
• Keywords: Code (set theory)

Utilizing infrastructure and vehicle-side information to track forecast the behaviors of surrounding traffic participants can significantly improve decision-making safety in autonomous driving. However, lack real-world sequential datasets limits research this area. To address issue, we introduce V2X-Seq, first large-scale V2X dataset, which includes data frames, trajectories, vector maps, lights captured from natural scenery. V2X-Seq comprises two parts: perception more than 15,000 frames 95 scenarios, trajectory forecasting contains about 80,000 infrastructure-view vehicle-view 50,000 cooperative-view scenarios 28 intersections' areas, covering 672 hours data. Based on three new tasks for vehicle-infrastructure cooperative (VIC) driving: VIC3D Tracking, Online-VIC Forecasting, Offline-VIC Forecasting. We also provide benchmarks introduced tasks. Find data, code, up-to-date at \href{https://github.com/AIR-THU/DAIR-V2X-Seq}{https://github.com/AIR-THU/DAIR-V2X-Seq}.