Are the tree based models being provided with data that's "kind of like" the data they were designed for/at training time for the deep learning model? Or "exactly like"?

I think that's perhaps the difference?