DLP

该数据集是一个公开的4K分辨率停车场人类驾驶视频数据集，具有精确的标注、高帧率以及丰富的交通场景。数据以图形结构组织，包含实例、代理、帧、障碍物和场景等元素，并可通过Python工具包实现查询和可视化。该数据集规模宏大，包含3.5小时的视频数据、30个场景、317,873帧画面、5,188个代理以及15,383,737个实例。其任务旨在进行车辆的多模态意图和轨迹预测。

This public 4K-resolution human driving video dataset is captured in parking lot environments, featuring precise annotations, high frame rates, and a diverse set of traffic scenarios. The dataset is structured in a graph format, including elements such as instances, agents, frames, obstacles, and scenarios, and supports querying and visualization via a dedicated Python toolkit. As a large-scale dataset, it contains 3.5 hours of video footage, 30 distinct scenarios, 317,873 frames, 5,188 agents, and 15,383,737 instances. The core downstream task supported by this dataset is multimodal intent and trajectory prediction for road vehicles.