in STN Computer Vision ~ read.

Deep Diffeomorphic Transformer Networks(CVPR2018)

Deep Diffeomorphic Transformer Networks(CVPR2018)
Proceedings: link
[Attention this paper have a supp, go to opeenacces.thecvf directly to get it!]

Idea:

  • instead of local linear A, they use the exp(A) instead so that we can introduce invertibility.

  • exp(A) can be consider as a Affine diffeomorphisms.

The exp(A) has closed form expression[You can find in supp], so the code is easy

Anyone can do conformal one?