China releases world-first open-source C-V2X dataset DAIR-V2X

Beijing (Gasgoo)- On February 24th, China released the world’s first C-V2X autonomous driving dataset based on real-life scenarios, the DAIR-V2X.

Photo credit: BJHAD

The DAIR-V2X is the first to achieve large-scale, multi-modality, multi-view synergy with 2D&3D object annotations. The dataset is co-released by the Institute for AI Industry Research, Tsinghua University (AIR), Beijing High-Level Autonomous Driving Demonstration Area, Beijing CheWang Technology Development Cooperation, Baidu Apollo, and Beijing Academy of Artificial Intelligence (BAAI). The dataset is available for download for users in China.

The data was collected from Beijing High-Level Autonomous Driving Demonstration Area in Beijing’s E-Town (Yizhuang District), consisting of 10 km city roads, 10 km highways, and 28 intersections. The data covers 71,254 image frames and 71,254 point-cloud frames transmitted from vehicle and road cameras, LiDARs, and other sensors. The data includes a diversity of scenarios under various weathers, including sunny, rainy, and cloudy days all day.

Aside from the multi-angle multimodal data and integrated annotations, this particular dataset adopts a semi-automatic self-learning C-V2X 3D integration method, cutting costs considerably.

The release of the DAIR-V2X is of great importance in China’s high-level autonomous driving development. The dataset has been included in BAAI’s open-source platform for the industry, academia, and research field to utilize and promote.