search

alibaba / Tora

The official repository for paper "Tora: Trajectory-oriented Diffusion Transformer for Video Generation"
https://ali-videoai.github.io/tora_video
Apache License 2.0
1.03k stars 38 forks source link

training data example #14

Open g-jing opened 19 hours ago

g-jing commented 19 hours ago

I understanding it is hard to release the full training data. Is it possible to release a training sample so we can follow it on customized data? Thanks!

zhenghao977 commented 14 hours ago

Hello, since the trajectories are extracted using the optical flow model during training, you only need to prepare the videos and text captions following the CogVideo5B format. We will release some examples as soon as possible.

Fancy93 commented 12 hours ago

Very interesting idea， why use the optical flow model for extracting trajectories?