SynVid
收藏资源简介:
SynVid是一个合成的视频数据集,由北京航空航天大学、上海人工智能实验室和香港大学共同创建。该数据集包含110000条高质量合成视频、去噪轨迹和对应的细粒度文本提示。SynVid利用教师模型生成高质量合成视频和去噪轨迹,确保了数据点的有效性和意义性。数据集的创建过程是基于教师模型在潜在空间中的操作,并通过精心设计的生成过程来生成。SynVid的应用领域主要是加速视频生成模型,解决视频生成过程中的速度和效率问题。
SynVid is a synthetic video dataset jointly created by Beihang University, Shanghai AI Laboratory, and The University of Hong Kong. This dataset contains 110,000 high-quality synthetic videos, denoising trajectories, and corresponding fine-grained text prompts. SynVid leverages teacher models to generate high-quality synthetic videos and denoising trajectories, ensuring the validity and meaningfulness of each data point. The dataset is constructed based on the operations of teacher models in the latent space, with the generation process carefully designed. The primary application scenario of SynVid is to accelerate video generation models, addressing the speed and efficiency issues encountered during video generation.




