search

ali-vilab / VGen

Official repo for VGen: a holistic video generation ecosystem for video generation building on diffusion models
https://i2vgen-xl.github.io
2.9k stars 255 forks source link

About the training dataset #5

Open Kevin-1342 opened 10 months ago

Kevin-1342 commented 10 months ago

Regarding the training dataset, would you mind me asking how did you collect 35 million single-shot text-video pairs from the long public datasets? My initial understanding was that the label for long video may not be suitable for short video clips. Many Thanks.

FuchenUSTC commented 9 months ago

The same concerns about the model training, is there any advices?