About training dataset

[RiverrHuang]

Thanks for your awesome work!
I was wondering what training dataset you used? The paper only mentions that it was a dataset collected from the internet, but I’d like to ask specifically: is it a paired dataset? And how was it obtained?

[johndpope]

probably best one
https://github.com/snap-research/EgoEdit/

there's ditto / edit 400gb