AFNI-sters:
I've read Cox and Jesmanowicz 99, reveled in the clever use of sheers to rotate etc, but what I really want is a qualitative understanding of how well it can be coaxed to behave in practice (to help us refine our use of it).
1. Qualitatively, what does this algorithm do with the edges, for example the edges of the volume, or the edges of a "-edging" box? If you set a box for the two volumes, then what magic avoids the algorithm simply aligning the boxes? Ie: why is it that the edges/sides of the box don't become the most salient feature for registration?
2. So I'm interested in the discussion of -weight option, and the corresponding "blurred sum" defaults. Perhaps the answer to my first question is that 3dvolreg applies some sort of fuzzy-edged weighting mask to taper off the contribution of voxels the closer they are to the edge?
2a. ... and if supplying a weighting brick, what range of values and data type are expected?
2b. Is there any way to extract the default weighting brick for inspection?
3. In several spots in the docs, users are cautioned that the algorithm is only good for a couple of voxels or couple of degress of registration. Big functional voxels or small anatomical voxels? Anyone care to report observations of how well it can do?
Thanks,
Graham