TED Gesture Dataset

该数据集包含了来自TED演讲的独白手势，共1,766个视频，这些视频经过处理以识别有效的手势动作及其对应的字幕。为了更精确地捕捉动作，该数据集还被重新处理成了SMPL-X姿态表示。规模上，数据集包含了1,766个视频，其任务是对语音进行手势合成。

This dataset comprises monologic gestures derived from TED Talks, encompassing a total of 1,766 videos. These videos have been processed to identify valid gestural motions and their corresponding subtitles. To achieve more precise motion capture, the dataset has been further reprocessed into SMPL-X pose representations. With a scale of 1,766 videos, its targeted task is speech-driven gesture synthesis.

搜集汇总

背景与挑战

背景概述

TED Gesture Dataset是一个包含1,766个TED演讲视频的数据集，平均每个视频时长为12.7分钟，包含35,685个感兴趣的镜头，总时长106.1小时。数据集提供了人体姿势提取、场景分割和转录对齐的脚本，适用于语音手势生成和深度学习研究。