Clarification of the arxiv paper content - Githubissues

YangLing0818 / VideoTetris

VideoTetris: Towards Compositional Text-To-Video Generation

https://arxiv.org/abs/2406.04277

MIT License

197 stars 6 forks source link

Clarification of the arxiv paper content #5

Open DachengLi1 opened 1 month ago

DachengLi1 commented 1 month ago

Thanks for the great work! Amazing results. I was looking at the arxiv paper and had some difficulty understand some key concepts. Could you kindly help me clarify?

I understand that (Section 3.1) given a written user prompt, you will decompose it spatially and locally, and compute attention score separately, and merge them. But I am not sure where is the autoregressive part. In this framework, how is the progressive following ability be achieved?

Thank you!