3DBench

3DBench是一个专为评估3D大型语言模型设计的可扩展基准数据集，由上海大学和上海AI实验室合作创建。该数据集包含超过23万个QA对，覆盖了从物体级到场景级的广泛空间和语义尺度，涉及感知和规划任务。数据集的创建过程包括从Procthor模拟框架中自动收集元数据，并利用这些数据生成多样化的3D对象和场景。3DBench的应用领域广泛，旨在解决当前多模态大型语言模型在3D理解方面的评估不足问题，推动相关技术的发展。

3DBench is a scalable benchmark dataset specifically designed for evaluating 3D large language models, jointly created by Shanghai University and Shanghai AI Laboratory. This dataset contains over 230,000 QA pairs, spanning a broad spectrum of spatial and semantic scales from object-level to scene-level, and covering perception and planning tasks. The development of 3DBench entails automatically collecting metadata from the Procthor simulation framework, and leveraging this metadata to generate diverse 3D objects and scenes. With a wide range of application scenarios, 3DBench aims to address the current gap in the evaluation of multimodal large language models for 3D understanding, and promote the advancement of related technologies.