We could add more descriptive error output in cases where we know the training will fail. For example, if the loaded network doesn't have the right number of volumes for the given output type (say 1 volume for 3d affinity output) we should explicitly tell that to the user.
This would also make it easier for us to diagnose mysterious errors, since we can rule out the obvious ones
We could add more descriptive error output in cases where we know the training will fail. For example, if the loaded network doesn't have the right number of volumes for the given output type (say 1 volume for 3d affinity output) we should explicitly tell that to the user.
This would also make it easier for us to diagnose mysterious errors, since we can rule out the obvious ones