一镜到底叙事视频数据集

采用长镜头连续拍摄方式进行视频采集，单条视频时长通常为数分钟以上。通过叙事设计将人物行为、环境变化与镜头运动进行连续表达。数据经过自动化检测筛选画面稳定性、曝光质量与音画同步，并对叙事阶段、人物行为、镜头运动进行结构化标签标注。

Video datasets are collected via long-take continuous shooting, with each individual video typically lasting several minutes or longer. Continuous presentation of character behaviors, environmental changes and camera movements is realized through narrative design. The collected data undergoes automated detection and screening based on frame stability, exposure quality and audio-video synchronization, and is then annotated with structured labels for narrative phases, character behaviors and camera movements.

搜集汇总

背景与挑战

背景概述

该数据集采用长镜头连续拍摄技术采集视频，单条视频时长通常为数分钟以上，通过叙事设计整合人物行为、环境变化和镜头运动。数据经过自动化检测筛选画面稳定性、曝光质量与音画同步，并对叙事阶段、人物行为、镜头运动进行结构化标签标注，适用于长视频理解、视频生成模型和叙事结构学习等场景。