METEOR

METEOR数据集是由马里兰大学计算机科学系创建，专注于捕捉非结构化场景中的交通模式和多代理驾驶行为。该数据集包含超过1000个一分钟的视频，超过200万个带边界框和GPS轨迹的标注帧，以及超过1300万个交通代理的边界框。METEOR数据集特别关注罕见和有趣的多代理驾驶行为，如交通违规、非典型交互和多样场景。每个视频都根据天气、一天中的时间、道路条件和交通密度等多种因素进行标记。该数据集用于评估对象检测和多代理行为预测的感知方法，特别适用于密集、异构和非结构化交通环境的研究。The METEOR Dataset was created by the Department of Computer Science, University of Maryland, and focuses on capturing traffic patterns and multi-agent driving behaviors in unstructured scenarios. This dataset contains over 1,000 one-minute videos, more than 2 million annotated frames with bounding boxes and GPS trajectories, as well as over 13 million bounding boxes for traffic agents. The METEOR Dataset specifically emphasizes rare and noteworthy multi-agent driving behaviors such as traffic violations, atypical interactions and diverse scenarios. Each video is labeled based on multiple factors including weather, time of day, road conditions and traffic density. This dataset is used to evaluate perception methods for object detection and multi-agent behavior prediction, and is particularly suitable for research on dense, heterogeneous and unstructured traffic environments.

搜集汇总

背景与挑战

背景概述

METEOR是一个大规模、高复杂性的交通数据集，专注于印度非结构化交通场景，包含超过100GB的视频数据、超过200万标注帧和1300万个边界框，注释涵盖罕见驾驶行为如切入、超速等。该数据集支持二维物体检测、行为动作预测和轨迹预测等研究任务，适用于计算机视觉和自动驾驶领域的模型开发与评估。