SPORTU
收藏资源简介:
SPORTU是一个综合性的体育理解基准数据集,由加州大学欧文分校等机构创建,旨在评估多模态大语言模型在体育领域的推理能力。数据集包含两个部分:SPORTU-text和SPORTU-video。SPORTU-text包含900个多选题,涵盖规则理解和策略分析,通过文本问答测试模型的推理能力;SPORTU-video则包含1,701个慢动作视频片段和12,048个问答对,评估从简单的体育识别到复杂的犯规检测和规则应用的多层次推理。数据集的创建过程包括人工注释和多角度视频采集,旨在解决体育分析和应用中的复杂推理问题。
SPORTU is a comprehensive sports understanding benchmark dataset created by institutions including the University of California, Irvine, aiming to evaluate the reasoning capabilities of multimodal large language models in the sports domain. The dataset consists of two components: SPORTU-text and SPORTU-video. SPORTU-text contains 900 multiple-choice questions covering rule comprehension and strategy analysis, which tests the model's reasoning ability through text-based question answering. SPORTU-video, on the other hand, includes 1,701 slow-motion video clips and 12,048 question-answer pairs, evaluating multi-level reasoning ranging from simple sports recognition to complex foul detection and rule application. The dataset's creation process involves manual annotation and multi-angle video collection, aiming to address complex reasoning challenges in sports analysis and applications.




