Content4All Open Research Sign Language Translation Datasets

Content4All Open Research Sign Language Translation Datasets是由英国萨里大学的视觉信号处理与模式识别中心等多个机构合作创建的大型数据集，包含190小时的新闻领域手语视频素材。数据集中的20小时视频由聋人专家和口译员进行了详细标注。数据集内容丰富，包括手语解释视频、对应的口语字幕以及提取的人体姿态特征。创建过程中，研究者们开发了专门的工具来对齐手语视频和字幕，并提供了基准翻译结果以支持未来的研究。该数据集主要应用于手语翻译研究，旨在解决当前手语研究中数据集规模小、领域受限的问题。Content4All Open Research Sign Language Translation Datasets is a large-scale dataset co-created by the Centre for Vision, Signal Processing and Pattern Recognition at the University of Surrey in the UK and multiple other institutions. It contains 190 hours of sign language video footage from the news domain. 20 hours of the dataset's videos have been meticulously annotated by deaf experts and interpreters. The dataset has rich content, including sign language interpretation videos, corresponding spoken subtitles, and extracted human pose features. During the creation process, researchers developed specialized tools to align sign language videos with subtitles, and provided benchmark translation results to support future research. This dataset is primarily applied to sign language translation research, aiming to address the issues of small dataset scale and limited domain in current sign language research.

搜集汇总

背景与挑战

背景概述

该数据集是Content4All项目发布的开放研究手语翻译数据集，旨在解决计算手语研究领域大规模数据缺乏的问题，推动实际应用发展。数据集包含六个子集，分为标注和原始两类，涵盖新闻和天气领域，提供视频、姿态估计数据、字幕和注释文件，需通过邮件申请获取，采用非商业性共享许可协议。