Large-scale Diverse Video (LDV) dataset

Large-scale Diverse Video (LDV)数据集由苏黎世联邦理工学院计算机视觉实验室创建，包含240个高分辨率视频，涵盖10种不同场景类别，如动物、城市、时尚等。数据集视频来源于YouTube，确保了高质量和无明显压缩瑕疵。创建过程中，视频经过降尺度处理以去除压缩瑕疵，并按标准格式转换为YUV 4:2:0。LDV数据集主要用于视频质量增强研究，特别是在压缩视频的质量提升方面，旨在通过提高压缩效率来改善视频的视觉体验和质量。

The Large-scale Diverse Video (LDV) dataset was developed by the Computer Vision Laboratory of ETH Zurich. It consists of 240 high-resolution videos covering 10 distinct scene categories including animals, urban scenes, fashion, and more. The videos are sourced from YouTube, guaranteeing high quality with no noticeable compression artifacts. During the dataset construction process, all videos were downscaled to remove compression artifacts and converted to the YUV 4:2:0 standard format. The LDV dataset is mainly intended for video quality enhancement research, especially for compressed video quality improvement, aiming to enhance visual experience and quality by optimizing compression efficiency.