DropletVideo-10M

该数据集包含了1000万个高质量视频，这些视频展现了动态的摄像机运动和物体动作，并附有详细描述各种摄像机移动和剧情发展的字幕注释。该数据集专注于时空变化，并精心筛选了一系列高质量视频片段，这些片段特别针对摄像机运动和物体动态进行了过滤。数据集规模宏大，包含1000万个视频，共计22.1亿帧，总视频时长达到20.4万小时。其任务重点在于生成具有时空一致性的视频。

This dataset contains 10 million high-quality videos featuring dynamic camera movements and object motions, accompanied by detailed subtitle annotations that elaborate on various camera maneuvers and plot developments. Focused on spatiotemporal variations, the dataset has been meticulously curated with high-quality video clips specifically filtered for camera motion and object dynamics. Boasting a massive scale, the dataset comprises 10 million videos, totaling 2.21 billion frames with an overall duration of 204,000 hours. Its core task focuses on generating videos with spatiotemporal consistency.