TVQA+

该数据集名为TVQA+，是首个融合了空间和时间注释的视频问答数据集，它基于原始的TVQA数据集构建。TVQA+包含了2,527个类别的注释，并支持三种任务：问答、时间定位和空间定位。该数据集的规模包括来自4,198个视频片段的29,383对问答，以及148,468张图像上标注的310,826个边界框。其任务重点在于进行时空视频问答。

This dataset, named TVQA+, is the first video question answering (QA) dataset that integrates both spatial and temporal annotations, and it is constructed based on the original TVQA dataset. TVQA+ includes annotations for 2,527 categories and supports three tasks: question answering, temporal localization, and spatial localization. In terms of dataset scale, it contains 29,383 QA pairs derived from 4,198 video clips, along with 310,826 bounding boxes annotated on 148,468 images. The core focus of its tasks is spatio-temporal video question answering.

搜集汇总

背景与挑战

背景概述

TVQA+是首个融合时空注释的视频问答数据集，基于原始TVQA构建，支持问答、时间定位和空间定位三种任务。该数据集包含来自4,198个视频片段的29,383对问答，并在148,468张图像上标注了310,826个边界框，旨在促进时空视频问答研究。