DAIR-V2X

DAIR-V2X数据集是第一个用于车路协同自动驾驶研究的大规模、多模态、多视图数据集。所有数据都是从真实场景中收集的，包括2D和3D注释。它具有以下特点:
-总共71254帧图像数据和71254帧点云数据；
-DAIR-V2X协作数据集 (DAIR-V2X-C)，包括38845帧图像数据和38845帧点云数据
-DAIR-V2X路边数据集 (DAIR-V2X-I)，包括10084帧图像数据和10084帧点云数据
-DAIR-V2X车辆数据集 (DAIR-V2X-V)，包括22325帧图像数据和22325帧点云数据
-首次实现了车路协同时空同步标注；
-传感器类型丰富，包括车侧摄像机，车侧激光雷达，路侧摄像机和路侧激光雷达；
-障碍物目标三维标注属性全面，标注道路上常见的10类障碍物目标；
-从北京高水平自动驾驶示范区的10公里城市道路、10公里高速公路、28个路口收集；
-数据涵盖了丰富的场景，例如晴天/雨天/雾天，白天/夜晚，城市道路/高速公路；
-数据完整，包括脱敏原始图像和点云数据，注释数据，时间戳，校准文件等。
我们根据5:2:3将数据集分为训练集、验证集和测试集。本次发布训练集和验证集，测试集和后续挑战活动一起发布。

DAIR-V2X is the first large-scale, multimodal, multi-view dataset for vehicle-infrastructure cooperative autonomous driving research. All data is collected from real-world scenarios, including 2D and 3D annotations. It has the following characteristics:
- 71,254 frames of image data and 71,254 frames of point cloud data in total;
- DAIR-V2X Cooperative Dataset (DAIR-V2X-C), consisting of 38,845 frames of image data and 38,845 frames of point cloud data;
- DAIR-V2X Infrastructure-side Dataset (DAIR-V2X-I), consisting of 10,084 frames of image data and 10,084 frames of point cloud data;
- DAIR-V2X Vehicle-side Dataset (DAIR-V2X-V), consisting of 22,325 frames of image data and 22,325 frames of point cloud data;
- It is the first dataset to provide spatio-temporally synchronized annotations for vehicle-infrastructure cooperative autonomous driving;
- It features a rich set of sensor modalities, including vehicle-side cameras, vehicle-side LiDARs, infrastructure-side cameras and infrastructure-side LiDARs;
- It provides comprehensive 3D annotation attributes for obstacle targets, covering 10 common categories of road-based obstacle targets;
- The data was collected from 10 km of urban roads, 10 km of expressways and 28 intersections within the high-level autonomous driving demonstration zone in Beijing;
- The dataset covers diverse scenarios, including sunny, rainy, foggy weather, daytime, nighttime, urban roads and expressways;
- The dataset is complete, containing anonymized raw image and point cloud data, annotation data, timestamps, calibration files and other relevant materials.
We split the entire dataset into training, validation and test sets at a ratio of 5:2:3. The training and validation sets are released in this version, while the test set will be released together with subsequent challenge activities.

AI搜集汇总

背景与挑战

背景概述

DAIR-V2X是首个用于车路协同自动驾驶研究的大规模真实场景数据集，包含超过7万帧图像和点云数据，分为协作、路边和车辆子集，支持多模态、多视图分析。数据集实现了车路协同时空同步标注，涵盖丰富传感器类型和10类障碍物目标，数据从北京城市道路、高速公路和路口收集，覆盖多种天气和光照条件。它由清华大学和百度于2022年发布，旨在推动自动驾驶协同感知技术的发展。