imregdemons vastly different displacement fields

I have two images, img01 and img02, and I am trying to generate the displacement fields between them using imregdemons. I had a script that did so perfectly. In this case my reference (fixed) image is img01, and the moving image is img02. I wrote a new script, using the same exact setup, and all of a sudden it isn't working. I know my images are different, and I know that I should be getting bigger displacements in the fields than I am currently getting. What's more, is when I switch it to do imregdemons(img01, img02) (so switching the moving and fixed), the displacements become WAY too big. Doing visual checks doesn't help - it verifies that the function isn't working properly. Any idea as to why switching the moving/fixed images changes everything so much, and anything I can do further to fix it?

[Alex Taylor]

The additive demons algorithm used by imregdemons is not guaranteed to be invertable, so the registrations of moving->fixed and fixed->moving can imply different displacement fields.

This is discussed in the following paper:

http://www-sop.inria.fr/asclepios/Publications/Tom.Vercauteren/DiffeoDemons-NeuroImage08-Vercauteren.pdf

I don't understand well enough the specifics of your data to comment on why different input images don't register well when a previous set of input images worked well.

[Martin Vaklev]

I had similar issues when the displacemnts between the two images get too large. Solved it by adding these parameters: imregdemons(...., 'PyramidLevels',10)