[Elastix] FW: Trouble with large BSpline registrations

Robin.Pearce at csiro.au Robin.Pearce at csiro.au
Thu Apr 9 01:30:44 CEST 2015


Hi Stefan,

I think I may have misread the output file, it's actually 20 million, not 2 billion, that seems to hit the limit. It's because I have an image which is about 2100x1500x1800 (~5.4GB) and I was using a grid spacing of 10 voxels in the final iteration. I was hoping to have a smaller grid spacing because there are small areas of the result which are badly deformed and I'm hoping to correct them.

For my purposes I don't actually need a full-on BSpline registration, as I'm looking at some before-and-after pictures of a solid, brittle material which has been compressed until breakage occurred, and I'm looking to put the pieces back together to look at where the cracking occurred. Essentially I need a piecewise rigid registration, where there are a number of large pieces which should line up with the original with just a rigid transformation. I'm not too keen on the idea of making one image for each piece, as that can get a little complicated for some samples, so I was hoping to use the BSpline registration, under the assumption that it should default to a piece-wise rigid registration, but I think my grid spacing is too large to allow that to work effectively. I'll have a go at limiting the number of jacobian measurements and let you know how I go.

Robin.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.bigr.nl/pipermail/elastix/attachments/20150408/4f20b82c/attachment.html>


More information about the Elastix mailing list