GePBench
收藏资源简介:
GePBench是由南京大学国家软件新技术重点实验室开发的一个新颖基准数据集,旨在评估多模态大语言模型在几何感知方面的能力。该数据集包含20,000张几何图形和250,000道多选题,涵盖了空间感知、形状理解等六个关键维度。数据集的构建过程通过专门的数据合成引擎生成结构化文本描述,并转化为几何图形,进而生成相应的多选题。GePBench的应用领域主要集中在提升多模态大语言模型在几何感知方面的基础能力,为复杂的视觉推理和决策任务奠定基础。
GePBench is a novel benchmark dataset developed by the State Key Laboratory for Novel Software Technology at Nanjing University, which aims to evaluate the geometric perception capabilities of multimodal large language models. This dataset contains 20,000 geometric graphics and 250,000 multiple-choice questions, covering six key dimensions including spatial perception and shape understanding. During the dataset construction process, a dedicated data synthesis engine is employed to generate structured text descriptions, which are then converted into geometric graphics, and subsequently corresponding multiple-choice questions are created. The application scenarios of GePBench mainly focus on enhancing the basic geometric perception capabilities of multimodal large language models, laying a solid foundation for complex visual reasoning and decision-making tasks.

- 1GePBench: Evaluating Fundamental Geometric Perception for Multimodal Large Language Models南京大学国家软件新技术重点实验室 · 2024年



