Diagnostic Evaluation of Video Inpainting on Landscapes (DEVIL) benchmark

DEVIL数据集是由密歇根大学创建的，专门用于视频修复的诊断评估。该数据集包含1250个视频片段，这些视频主要来自Flickr上的风景视频，经过筛选和处理，确保不包含前景对象。数据集通过标注视频和遮罩的多种关键内容属性，如相机运动和背景场景运动，来帮助分析视频修复方法的性能。DEVIL数据集的应用领域主要集中在视频修复技术，旨在通过详细的属性分析，提高视频修复的质量和真实性。

The DEVIL dataset was created by the University of Michigan specifically for diagnostic evaluation of video inpainting. It includes 1,250 video clips primarily sourced from landscape videos on Flickr, which have been screened and processed to ensure that no foreground objects are included. By annotating multiple key content attributes of the videos and their corresponding masks, such as camera motion and background scene motion, the dataset facilitates performance analysis of video inpainting methods. The application scope of the DEVIL dataset is mainly focused on video inpainting technologies, aiming to improve the quality and realism of video inpainting through detailed attribute analysis.