SHIFT
收藏资源简介:
SHIFT是一个由苏黎世联邦理工学院创建的大型合成驾驶数据集,专注于连续多任务域自适应。该数据集包含4850个序列,覆盖了从云量、雨雾强度到时间和车辆行人密度的多种环境变化。SHIFT数据集通过其全面的传感器套件和主流感知任务的标注,允许研究者探索感知系统性能在域移位增加时的退化情况,从而推动连续适应策略的发展,评估模型鲁棒性和通用性。数据集的应用领域包括自动驾驶中的域泛化、域适应和不确定性估计,旨在解决自动驾驶系统在不断变化环境中的安全性和适应性问题。
SHIFT is a large-scale synthetic driving dataset developed by ETH Zurich, focusing on continuous multi-task domain adaptation. This dataset comprises 4850 sequences, covering a diverse set of environmental variations spanning cloud cover, rain/fog intensity, time of day, as well as the densities of vehicles and pedestrians. Equipped with a comprehensive sensor suite and annotations for mainstream perception tasks, the SHIFT dataset enables researchers to investigate the degradation of perception system performance with intensified domain shift, thereby facilitating the advancement of continuous adaptation strategies and supporting the evaluation of model robustness and generalizability. The application scenarios of this dataset include domain generalization, domain adaptation and uncertainty estimation in autonomous driving, with the goal of addressing the safety and adaptability issues faced by autonomous driving systems in dynamically changing environments.



