NExT-QA

NExT-QA 共包含 5440 个视频，平均长度为 44 秒，约 52K 手动注释的问答对，分为因果（48%）、时间（29%）和描述性（23%）问题。我们为每个视频特别注释了大约 10 个问题，涵盖了不同的内容。数据集分为train/val/test: 3870/570/1000，其中发布了1000个测试视频。

NExT-QA contains a total of 5,440 videos with an average duration of 44 seconds, and approximately 52K manually annotated question-answer pairs. These pairs are categorized into three types: causal (48%), temporal (29%), and descriptive (23%). We specifically annotated around 10 questions per video to cover diverse content. The dataset is split into training, validation and test sets with sizes of 3,870, 570 and 1,000 respectively, where 1,000 test videos have been released.

搜集汇总

背景与挑战

背景概述

NExT-QA是一个视频问答数据集，包含5440个平均44秒的视频和约52K手动注释的问答对，问题类型以因果、时间和描述性为主，旨在支持解释性时间动作分析。数据集分为训练、验证和测试集，其中测试集包含1000个已发布的视频，由新加坡国立大学于2021年发布。