jcreinhold/synthtorch

Full image coverage in patch-based 3D prediction

jcreinhold opened this issue · 2 comments

The 3D patch-based prediction routine is poorly written and needs to be more thoroughly tested.

Current setup requires that the patch-size be a GCD of the image shape to get full coverage. However, patches do appear to be gathered correctly (i.e., in the same orientation as gathered in training).

Fixed with dec6f5f