I don't think that's true. If they masked out the backgrounds on training, how would the model be able to synthesize a background at all? The problem is there'

I don't think that's true. If they masked out the backgrounds on training, how would the model be able to synthesize a background at all? The problem is there's too much variety in the backgrounds of the training set. They don't follow a pattern the way growing a human does.