MUSES
收藏资源简介:
MUSES数据集是由苏黎世联邦理工学院创建的多传感器语义感知数据集,专为在不确定性下驾驶设计。该数据集包含2500个样本,每个样本包括同步的多模态记录,如2D全景标注的图像,以及正常帧摄像头、激光雷达、雷达、事件摄像头和IMU/GNSS传感器的读数。数据集涵盖了多种天气和光照条件,旨在支持模型训练和评估在多样视觉条件下的性能。MUSES数据集的标注过程采用两阶段协议,能够捕捉类别和实例级别的不确定性,从而支持不确定性感知的全景分割任务,这是该数据集的一个创新点。此外,数据集的应用领域包括多模态密集语义感知研究,以及探索不同传感器在复杂环境下的表现。
The MUSES dataset is a multi-sensor semantic perception dataset created by ETH Zurich, specifically designed for driving under uncertainty. It contains 2500 samples, each of which includes synchronized multimodal recordings, such as 2D panoptically annotated images, as well as readings from regular frame cameras, LiDAR, radar, event cameras, and IMU/GNSS sensors. The dataset covers a variety of weather and lighting conditions, aiming to support model training and performance evaluation under diverse visual conditions. The annotation process of the MUSES dataset adopts a two-stage protocol, which can capture category-level and instance-level uncertainties, thereby supporting uncertainty-aware panoptic segmentation tasks—this is an innovative feature of the dataset. In addition, the application areas of the dataset include multimodal dense semantic perception research, and the exploration of the performance of different sensors in complex environments.




