Koala-36M
收藏资源简介:
Koala-36M是由快手科技创建的大规模高质量视频数据集,旨在提升视频生成模型的一致性。该数据集包含3600万条视频,平均每条视频时长为13.75秒,分辨率为720p,每条视频配有平均202字的详细描述。数据集的创建过程包括精确的时间分割、详细的字幕生成和高质量视频过滤。Koala-36M主要应用于视频生成领域,旨在通过提高文本与视频内容的一致性来优化生成模型的性能。
Koala-36M is a large-scale high-quality video dataset created by Kuaishou Technology, aiming to enhance the consistency of video generation models. This dataset contains 36 million videos, with an average duration of 13.75 seconds per video, a resolution of 720p, and each video is accompanied by a detailed caption averaging 202 words. The dataset's creation process includes precise temporal segmentation, detailed caption generation, and high-quality video filtering. Koala-36M is primarily applied in the field of video generation, with the goal of optimizing the performance of generation models by improving the consistency between text and video content.




