You seem to be losing a battle with truncation errors.
There are 5 datasets with a minimum dimension less than
3.0, but they are 2.999999 or 2.999998. It may be a bit
irritating, but add '-volreg_warp_dxyz 3' to the
afni_proc.py commands, to be sure they all come out as
3.0.
Currently, afni_proc.py uses the smallest dimension
truncated to 3 significant bits. So anything less than
3 drops to 2.5. But I have seen other cases where the
dimension is just below a useful number, so I will change
the method to
first: round to 6 significant bits
next: truncate to 3
So if the dimension is within 6 sig bits of the next
higher change of 3 sig bits, it will round up, instead.
Note that we generally don't want to round up and make
voxels bigger than acquired. But this seems like a good
compromise.
That change should be available tonight.
- rick
