MUAD
收藏资源简介:
MUAD数据集是由巴黎综合理工学院创建的,专注于自动驾驶中多种不确定性的评估。该数据集包含10,413张合成图像,涵盖了夜间、雾、雨、雪等多样化的不利天气条件,以及分布外对象,用于语义分割、深度估计、对象和实例检测的标注。MUAD旨在更好地评估模型在不同不确定性源下的性能,特别是在自动驾驶场景中,通过提供详细的基准和评估工具,推动深度神经网络在敏感应用中的可靠性研究。
The MUAD dataset was developed by École Polytechnique, with the core goal of evaluating multiple uncertainties in autonomous driving. It consists of 10,413 synthetic images covering diverse adverse weather conditions including nighttime, fog, rain and snow, as well as out-of-distribution objects, and is annotated for tasks including semantic segmentation, depth estimation, object detection and instance detection. The MUAD dataset aims to better assess model performance across different sources of uncertainty, particularly in autonomous driving scenarios, and advance reliability research for deep neural networks in safety-critical applications by offering comprehensive benchmarks and evaluation tools.




