OK, I have a new theory about what's going on. I can take an anatomical +tlrc dataset with both .HEAD and .BRIK files and no associated +orig or +acpc bits, and use it with adwarp to produce a Talairach'd functional dataset from a +orig functional dataset. This process yields perfectly nice values.
Unfortunately when I translate the +tlrc anatomical dataset to Pittsburgh MRI and back to AFNI using our tools, and then use that translated dataset as input to adwarp, the output of adwarp is all zeros. In the process of translation, the data gets reordered from RAS to RAI and appropriate changes get made to the dataset geometry info. The untranslated and translated datasets look identical in afni and to 3dinfo, but somehow they are not identical.
I don't change the WARP_DATA attribute, however- I assumed it was written in some absolute coordinate system and thus didn't depend on the RAS/RAI business. I guess I do have to change the WARP_DATA. Does that sound correct? Any hints would be much appreciated.
Thanks,
-Joel