Training a 3D model #19
Description
Given that the ground truths are single 2D slices, it is not possible to directly train a 3D model. However, one solution would be to get a 3D ground truth using a first pass of a 2D model, and use that inference output as the ground truth for a 3D model.
Other related approach, is to train a 2.5D model using adjacent slices as multi-channel input #14