A Survey on Autonomous Driving Datasets: Statistics, Annotation Quality, and a Future Outlook

A Survey on Autonomous Driving Datasets: Statistics, Annotation Quality, and a Future Outlook

23 Apr 2024 | Mingyu Liu*, Ekim Yurtsever, Member, IEEE, Jonathan Fossaert, Xingcheng Zhou, Walter Zimmer, Yuning Cui, Student Member, IEEE, Bare Luka Zagar, Alois C. Knoll, Fellow, IEEE
This paper presents an extensive survey of 265 autonomous driving datasets, covering various aspects such as sensor modalities, data size, tasks, and contextual conditions. The authors introduce a novel metric, the impact score, to evaluate the significance of datasets, which can also guide the creation of new datasets. The survey analyzes the annotation processes, existing labeling tools, and annotation quality, emphasizing the importance of a standardized annotation pipeline. It also examines the impact of geographical and adversarial environmental conditions on autonomous driving systems' performance. The paper discusses the data distribution of several key datasets and their pros and cons, and concludes with a discussion on current challenges and future trends in autonomous driving datasets. The main contributions include a comprehensive overview of datasets, a detailed analysis of sensor modalities, and an in-depth look at the tasks and challenges in autonomous driving. The survey aims to facilitate the development of future algorithms and datasets in the field of autonomous driving.This paper presents an extensive survey of 265 autonomous driving datasets, covering various aspects such as sensor modalities, data size, tasks, and contextual conditions. The authors introduce a novel metric, the impact score, to evaluate the significance of datasets, which can also guide the creation of new datasets. The survey analyzes the annotation processes, existing labeling tools, and annotation quality, emphasizing the importance of a standardized annotation pipeline. It also examines the impact of geographical and adversarial environmental conditions on autonomous driving systems' performance. The paper discusses the data distribution of several key datasets and their pros and cons, and concludes with a discussion on current challenges and future trends in autonomous driving datasets. The main contributions include a comprehensive overview of datasets, a detailed analysis of sensor modalities, and an in-depth look at the tasks and challenges in autonomous driving. The survey aims to facilitate the development of future algorithms and datasets in the field of autonomous driving.
Reach us at info@study.space