JRDB-PanoTrack 2024 - 开放世界全景分割和跟踪数据集

JRDB-PanoTrack数据集由蒙纳士大学、穆罕默德·本·扎耶德人工智能大学和阿德莱德大学联合构建，是一个旨在为机器人和视觉应用设计的开放世界全景分割和跟踪数据集。该数据集包含了精细标注的428,000张全景分割图和27,000条掩模轨迹，涵盖71个对象类别，分为60个thing类（例如行人、汽车、笔记本电脑）和11个stuff类（如天空和墙壁）。数据集的创建过程始于对原始JRDB数据集的扩展，通过合并不同视角的图像生成全景图像，并提供相应的点云数据。注释过程由专业注释者执行，确保了数据的质量和多样性。JRDB-PanoTrack数据集为机器人在人类密集环境中的自主性研究提供了新的基准和挑战，推动了环境理解、物体分割和多目标追踪技术的发展。

JRDB-PanoTrack dataset was jointly constructed by Monash University, Mohamed bin Zayed University of Artificial Intelligence, and The University of Adelaide. It is an open-world panoptic segmentation and tracking dataset designed for robotic and computer vision applications. The dataset contains 428,000 finely annotated panoramic segmentation maps and 27,000 mask tracks, covering 71 object categories, which are divided into 60 thing classes (e.g., pedestrians, cars, laptops) and 11 stuff classes (e.g., sky and walls). The dataset was developed by expanding the original JRDB dataset, generating panoramic images by merging images from different perspectives, and providing corresponding point cloud data. The annotation process was carried out by professional annotators to ensure the quality and diversity of the dataset. The JRDB-PanoTrack dataset offers a new benchmark and challenge for autonomous robotics research in human-dense environments, and promotes the advancement of environment understanding, object segmentation, and multi-object tracking technologies.