iVISPAR
收藏资源简介:
iVISPAR是一个互动式的视觉空间推理基准,由奥斯纳布吕克大学认知科学研究所创建,旨在评估视觉语言模型在动态环境中的空间推理能力。该数据集基于滑动拼图游戏,支持2D、3D和基于文本的输入模态,包含了从简单配置到NP完全挑战的不同难度级别的任务。数据集通过调整板块大小、方块数量和解决方案路径等因素来实现任务复杂性的精细调整,为研究提供了评估视觉语言模型空间推理和规划能力的坚实基础。
iVISPAR is an interactive visual spatial reasoning benchmark developed by the Institute of Cognitive Science at the University of Osnabrück. It is designed to evaluate the spatial reasoning capabilities of vision-language models (VLMs) in dynamic environments. Built upon sliding puzzle games, the benchmark supports three input modalities: 2D, 3D, and text-based inputs, and encompasses tasks across a wide range of difficulty levels, from simple configurations to NP-complete challenges. The complexity of each task can be finely tuned by adjusting factors such as tile size, number of blocks, and solution paths, thus providing a solid foundation for researching and assessing the spatial reasoning and planning abilities of vision-language models.




