Panda-70M

该数据集是一组用于训练视频扩散模型的Ctrl-Adapter的大规模视频训练集，已经过筛选，去除了静态场景。此外，为了提高数据质量，还过滤掉了平均光流低于0.25的视频，最终筛选后保留了大约200万个视频。这些视频将用于视频生成模型的训练任务。

This dataset is a large-scale video training set for training the Ctrl-Adapter of video diffusion models. Static scenes have been removed via screening. Furthermore, to enhance data quality, videos with an average optical flow lower than 0.25 were also filtered out. Approximately 2 million videos were retained after the final screening, and these videos will be used for the training of video generation models.