HYouTube

HYouTube数据集由上海交通大学创建，专注于视频和谐化领域，旨在通过调整视频前景使其与背景在色彩和光照上兼容。该数据集基于YouTubeVOS 2018构建，包含3194对合成复合视频和真实视频，每对视频包含20帧。创建过程中，使用了超过400种3D颜色查找表（LUT）来调整前景外观，确保视频多样性。HYouTube数据集主要用于解决视频和谐化问题，提升合成视频的真实感。

The HYouTube dataset was developed by Shanghai Jiao Tong University, focusing on the domain of video harmonization. Its core objective is to adjust the foreground of videos to achieve color and lighting compatibility with the corresponding background. Built upon YouTubeVOS 2018, the dataset encompasses 3,194 pairs of synthetic composite video and real video samples, where each sample contains 20 frames. During the dataset creation process, more than 400 types of 3D color lookup tables (LUTs) were utilized to adjust the foreground appearance, thereby guaranteeing the diversity of the video samples. The HYouTube dataset is primarily intended to solve the video harmonization problem and improve the realism of synthetic composite videos.