furonghuang-lab/morse-500

MORSE-500数据集包含500个新颖的视频片段，这些视频是程序化生成的，涵盖了抽象、数学、物理、计划、空间和时间六个推理类别，用于测试模型的多模态推理能力。数据集具有可扩展难度，可以根据模型的改进发布更难的版本。视频问题直接嵌入在视频中，要求模型不依赖文本信息进行推理。数据集提供了开发者友好的浏览和调试工具。

The MORSE-500 dataset consists of 500 novel video clips spanning six reasoning categories: abstract, mathematical, physical, planning, spatial, and temporal, designed to benchmark multimodal reasoning capabilities. The dataset features scalable difficulty, allowing for the release of harder versions as models improve. The videos embed questions directly, requiring models to perform reasoning without relying on text cues. Developer-friendly tools for browsing and debugging are provided.