PST900

PST900数据集由宾夕法尼亚大学的GRASP实验室创建，包含894对同步和校准的RGB和热图像，每像素都有人类注释，涵盖四个不同的类别。数据集来源于DARPA地下挑战赛，旨在解决在光照和可见度受限的环境中，如地下隧道、矿井和洞穴中的语义分割问题。数据集的创建过程涉及在多种环境中收集代表性数据，并由实验室内的标注人员进行标注和验证。PST900数据集的应用领域主要集中在机器人自主性和环境感知，特别是在极端环境下的物体识别和跟踪。

The PST900 dataset was created by the GRASP Laboratory at the University of Pennsylvania. It contains 894 pairs of synchronously captured and calibrated RGB and thermal images, with human annotations at the pixel level, covering four distinct categories. Derived from the DARPA Subterranean Challenge, this dataset is designed to address semantic segmentation challenges in environments with limited lighting and visibility, such as underground tunnels, mines, and caves. The dataset's creation involved collecting representative data across diverse environments, followed by annotation and validation performed by in-house annotators. The primary application scenarios of the PST900 dataset focus on robotic autonomy and environmental perception, particularly object recognition and tracking in extreme environments.