3D2T-SR, 3D2T-NR, 3D2T-QA

本研究构建了三个新的基准数据集3D2T-SR、3D2T-NR和3D2T-QA，用于点云与文本匹配任务。这些数据集包含全面的场景描述，旨在评估点云与文本匹配的性能。数据集的创建过程涉及使用大型语言模型生成描述，并通过多次迭代确保描述的质量和准确性。这些数据集适用于室内/城市峡谷定位和场景检索等应用，旨在解决点云与文本间精确匹配的问题。

This study constructs three novel benchmark datasets, namely 3D2T-SR, 3D2T-NR and 3D2T-QA, for point cloud-text matching tasks. These datasets feature comprehensive scene descriptions, designed to evaluate the performance of point cloud-text matching systems. The development of these datasets involves generating descriptions using Large Language Models (LLMs), and iteratively refining the descriptions to ensure their quality and accuracy. These datasets are applicable to applications such as indoor/urban canyon localization and scene retrieval, aiming to address the challenge of precise matching between point clouds and text.