Closed jcreinhold closed 5 years ago
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 dec6f5f106dc0d98027b99738cb5e35857de7a0d
The 3D patch-based prediction routine is poorly written and needs to be more thoroughly tested.