Genesis_VQA

本数据集是一个大规模的多模态视频理解与推理基准数据集, 视频内容涵盖多种现实生活场景（辩论、家庭、教育、社交、军事、历史等），平均时长约50分钟。数据集专注于评估模型在长视频场景下的综合理解能力，从基础的视觉感知到高层次的逻辑推理。This dataset is a large-scale multimodal video understanding and reasoning benchmark dataset. Its video content covers various real-life scenarios including debates, family, education, social interactions, military affairs, history, etc., with an average duration of approximately 50 minutes. This dataset focuses on evaluating the comprehensive understanding capabilities of models in long-video scenarios, ranging from basic visual perception to high-level logical reasoning.

搜集汇总

背景与挑战

背景概述

Genesis_VQA是一个大规模多模态视频理解与推理基准数据集，包含346个平均时长约50分钟的视频，覆盖现实生活多种场景。该数据集通过8类问题类型（共约5500多个问题）系统评估模型在长视频中的综合理解能力，采用多选题形式并配备精确时间标注。