I just stumbled on something weird that may be worth to report.

With 3dQwarp, I non-linearly register two EPI images that were acquired with the same parameters except for the direction of phase encoding (for EPI distortion correction), and the output images are as expected half-way between the input images:

3dQwarp -overwrite -prefix funMean.nii.gz -base distRef.nii.gz -weight distRef-mask.nii.gz -source funMean.nii.gz -noZdis -noXdis -plusminus -superhard -minpatch 9 -nmi

Applying the inverse of the displacement map to the output (the undistorted image) in order to recover the input (the distorted image) gives weird results. The original and recovered distorted images don't match. Parts of the image are displacement in direction other then the one it is supposed to be displaced:

3dNwarpApply -overwrite -prefix funMeanX.nii.gz -nwarp 'INV(funMean_PLUS_WARP.nii.gz)' -source funMean_PLUS.nii.gz

But inverting the displacement map sperately before applying it worked!?:
3dNwarpCat -prefix funMean_PLUS_WARPX.nii.gz -warp1 'INV(funMean_PLUS_WARP.nii.gz)'
3dNwarpApply -overwrite -prefix funMeanXX.nii.gz -nwarp funMean_PLUS_WARPX.nii.gz -source funMean_PLUS.nii.gz

Is that some bug or a silly mistake I am making?

Thanks a lot!
Sébastien