You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Deformable convolutions (paper, implementation) might be an interesting feature that could potentially improve invariance against geometric deformations.
I am not sure yet how deformable convolutions can be used as a (perhaps inferior) alternative to our image warping pipeline or if they can reasonably work together and improve general robustness of trained networks.
The text was updated successfully, but these errors were encountered:
Deformable convolutions (paper, implementation) might be an interesting feature that could potentially improve invariance against geometric deformations.
I am not sure yet how deformable convolutions can be used as a (perhaps inferior) alternative to our image warping pipeline or if they can reasonably work together and improve general robustness of trained networks.
The text was updated successfully, but these errors were encountered: