Charades-Ego
收藏资源简介:
Charades-Ego数据集由卡内基梅隆大学创建,是一个包含68,536个活动实例的大型数据集,涵盖68.8小时的第三和第一人称视频。数据集通过网络招募的众包工作者录制,采用‘好莱坞在家’方法,确保了数据的多模态和多样性。数据集内容包括从第三视角到第一视角的活动同步记录,以及文本描述和时间标注,适用于视频分类、定位和字幕生成等任务。Charades-Ego旨在通过结合第三和第一人称视频理解,提升增强现实和虚拟现实等应用中的活动识别能力。
Created by Carnegie Mellon University, the Charades-Ego dataset is a large-scale dataset containing 68,536 activity instances and spanning 68.8 hours of third-person and first-person video footage. Recorded by online-recruited crowdworkers using the "Hollywood in Homes" methodology, this dataset ensures the multimodality and diversity of its contents. It includes synchronized activity recordings from third-person to first-person perspectives, alongside textual descriptions and temporal annotations, making it suitable for tasks such as video classification, localization, and video caption generation. The Charades-Ego dataset aims to improve activity recognition capabilities in applications like augmented reality (AR) and virtual reality (VR) by combining third-person and first-person video understanding.

- 1Charades-Ego: A Large-Scale Dataset of Paired Third and First Person Videos卡内基梅隆大学 · 2018年



