There's one paper I found that explores this approach: https://arxiv.org/pdf/1708.02731.pdf

I know little about deep learning though, so I can't really comment on it.

For situations like the roof problem you describe, I think PatchMatch is more likely to work out well. The downside is that it's considerably harder to implement correctly than seam carving. http://gfx.cs.princeton.edu/pubs/Barnes_2009_PAR/patchmatch....