Sports-QA
收藏资源简介:
Sports-QA数据集是专为体育视频问答任务设计的大型数据集,包含约94,000个问答对,涵盖多种体育项目。该数据集通过从MultiSports和FineGym数据集中提取高质量体育视频和专业动作标签构建而成。数据集中的问题类型多样,包括描述性、时间性、因果性和反事实条件问题,旨在评估模型对专业体育动作的理解和精细运动分析的能力。Sports-QA数据集不仅适用于视频问答任务,还可用于多标签分类和体育视频字幕生成等其他任务,为体育视频理解领域提供了全面的评估平台。
Sports-QA Dataset is a large-scale dataset specifically developed for sports video question answering (Video QA) tasks, containing approximately 94,000 question-answer pairs covering a diverse array of sports disciplines. This dataset is constructed by extracting high-quality sports videos and professional action labels from the MultiSports and FineGym datasets. The dataset includes diverse question types, such as descriptive, temporal, causal, and counterfactual conditional questions, aiming to evaluate a model's understanding of professional sports movements and its ability for fine-grained motion analysis. Beyond video QA tasks, the Sports-QA Dataset can also be applied to other tasks including multi-label classification and sports video caption generation, providing a comprehensive evaluation platform for the field of sports video understanding.




