GarrickAI/VideoMarathon

VideoMarathon是一个大规模的长视频指令跟随数据集，总时长约为9,700小时，包含330万个问答对，覆盖22个任务类别。这些任务涉及六个基本主题：时间性、空间性、对象、动作、场景和事件，要求进行短形式和长形式的视频理解。数据集源自多样化的视频源领域，视频长度从三分钟到一小时不等，包含复杂的事件计数内容。数据样本包括唯一ID、视频源、视频路径、URL、问题类型（如多选或开放性问题）、问题和答案，适用于学术研究目的。VideoMarathon is a large-scale long video instruction-following dataset with a total duration of approximately 9,700 hours, comprising 3.3 million QA pairs across 22 task categories. These tasks span six fundamental topics: temporality, spatiality, object, action, scene, and event, requiring both short-form and long-form video comprehension. The dataset draws from diverse video source domains, with video durations ranging from three minutes to one hour, and includes complex event counting. Data samples feature a unique ID, data source, video path, URL, question type (e.g., multiple-choice or open-ended), question, and answer, intended for academic research purposes only.