search

ali-vilab / UniAnimate

Code for Paper "UniAnimate: Taming Unified Video Diffusion Models for Consistent Human Image Animation".
https://unianimate.github.io/
982 stars 52 forks source link

Details of your video #25

Open xiaohutongxue-sunny opened 3 months ago

xiaohutongxue-sunny commented 3 months ago

Great and amazing work! Thanks for your contribution, would you mind tell more details about your data, videos need a fixed background or not? Is there any other train tricks can be shared or some suggestions for fine-tuning on our own data about 1600 videos from internet. Thank you.

xiaohutongxue-sunny commented 3 months ago

Do you use additional pre-process for your data, such as extracting objects and replacing background with green canvas

wangxiang1230 commented 3 months ago

Hi， we use videos with less background movements. And preprocess you mentioned is not used. Thanks.

xiaohutongxue-sunny commented 3 months ago

Hi， we use videos with less background movements. And preprocess you mentioned is not used. Thanks.

Thanks for your replay，is there some preprocess ways were used for your data?