tomg-group-umd/morse-500

MORSE-500 Benchmark是一个包含500个程序生成的视频的数据集，用于测试多模态推理能力。这些视频涵盖六个推理类别：抽象、数学、物理、规划、空间和时间。数据集设计新颖，难度可扩展，且不含文本辅助，纯粹基于视觉推理。数据集提供快速运行的CSV元数据和MP4视频文件，支持开发者便捷地浏览和调试。

MORSE-500 Benchmark is a dataset consisting of 500 programatically generated videos designed to stress-test multimodal reasoning capabilities. These videos span across six reasoning categories: abstract, mathematical, physical, planning, spatial, and temporal. The dataset is novel, has scalable difficulty, and relies purely on visual reasoning without text aids. It provides CSV metadata and MP4 video files for fast runtime and developer-friendly browsing and debugging.