Bench2Drive
收藏资源简介:
Bench2Drive是由上海交通大学人工智能与计算机科学与工程学院创建的一个大型数据集,包含200万个完全标注的帧,来源于10000个短片段,均匀分布在44个交互场景、23种天气和12个城镇中。该数据集通过CARLA v2收集,旨在为全自动驾驶(FSD)提供一个全面、真实和公平的测试环境。创建过程中,数据集采用了先进的专家模型Think2Drive进行数据收集和标注。Bench2Drive的应用领域广泛,主要用于评估端到端自动驾驶(E2E-AD)系统的多重能力,解决现有评估方法中存在的不足,如开放环路评估的局限性和闭环评估中的高变异性。
Bench2Drive is a large-scale dataset developed by the School of Artificial Intelligence and Computer Science and Engineering, Shanghai Jiao Tong University. It contains 2 million fully annotated frames derived from 10,000 short clips, which are uniformly distributed across 44 interactive driving scenarios, 23 weather conditions, and 12 towns. Collected via CARLA v2, this dataset is designed to offer a comprehensive, realistic and fair testing environment for fully autonomous driving (FSD). During its construction, the advanced expert model Think2Drive was utilized for data collection and annotation. Bench2Drive has broad application domains, and is primarily employed to evaluate the multiple capabilities of end-to-end autonomous driving (E2E-AD) systems, while addressing the drawbacks of existing evaluation methods including the limitations of open-loop evaluation and the high variability present in closed-loop evaluation.




