Text-to-Video Person Re-identification (TVPReid) dataset
收藏arXiv2024-02-02 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2307.07184v2
下载链接
链接失效反馈官方服务：
资源简介：
TVPReid数据集由南京工业大学创建，包含1734个视频，总大小为2.06G，源自PRID2011和iLIDS-VID两个现有数据集。每个视频配有两个不同的文本描述，共计3468个描述句，涵盖多种场景、视角和摄像机规格，增强了视频内容的多样性。数据集创建过程中，通过OpenCV合成了视频，并去除了相似样本。TVPReid数据集主要用于文本到视频的人物检索研究，旨在通过视频中的连续帧提供更丰富的动态信息，解决文本到图像人物检索中因孤立帧导致的性能限制问题。
The TVPReid dataset, developed by Nanjing Tech University, consists of 1734 videos with a total size of 2.06 GB, and is sourced from two existing datasets: PRID2011 and iLIDS-VID. Each video is paired with two unique textual descriptions, yielding a total of 3468 descriptive sentences. The dataset covers diverse scenarios, viewpoints and camera specifications, which further enhances the diversity of video content. During the dataset construction process, videos were synthesized using OpenCV, and similar samples were removed. Primarily intended for text-to-video person retrieval research, the TVPReid dataset aims to provide richer dynamic information via consecutive frames in videos, so as to address the performance limitations caused by isolated frames in text-to-image person retrieval.
提供机构：
南京工业大学
创建时间：
2023-07-14
