MSVD-QA

该数据集名为MSVD-QA，是基于原始MSVD数据集生成的开放式视频问答数据集，其中包含了视频以及自动生成的问答标注对。问题被分为五类：分别是“什么”、“谁”、“何时”、“哪里”和“怎样”，而答案则通常是一词名词或动词。该数据集规模包括1.9千个视频和5万个问答对，其任务是进行视频问答。

The dataset named MSVD-QA is an open-ended video question answering (VideoQA) dataset derived from the original MSVD dataset. It includes videos and automatically generated question-answer annotation pairs. Questions are divided into five categories: "what", "who", "when", "where", and "how", while answers are typically single-word nouns or verbs. This dataset comprises 1,900 videos and 50,000 question-answer pairs, and its core task is video question answering.

搜集汇总

背景与挑战

背景概述

MSVD-QA是一个基于MSVD视频描述数据集生成的视频问答数据集，包含从视频描述中生成的问答对，适用于视频问答任务的研究和开发。