ecr05
commented
6 years ago Hi Roberto,
You must increase regularisation, not decrease. Not knowing your data it is difficult to be confident in what values to suggest, but try running 0,0.3,0.4,0.5. Generally I optimise by running one resolution level at a time using the --levels parameter so run --levels=2 (as 2 is your first discrete level) with different regularisation parameters for the second level i.e. currently its 0.1, try 0.2, 0.3, 0.4 ... 0.5 then move onto the third level.
The regularisation in msm_hocr is much more robust, and will lead to better alignment with lower distortions, particularly for noisy data such as fMRI. However, if you are just working with folding patterns I think the FSL version will work well for you - just increase the regularisation somewhat. Note, the optimisation used in the FSL version means it is also much faster to run.
Let me know if you need more help
Thanks
Emma
On 9 January 2018 at 12:00, Roberto Toro notifications@github.com wrote:
Hello! Thank you for MSM : ) I'm working with the version that comes with FSL, and although it often works very nicely, sometimes there are regions that get veery large deformations. Is there a way of limiting the maximum amount of deformation allowed? I tried lowering down the lambdas, from 0,0.1,0.2,0.3 to 0,0.01,0.02,0.03, but I still get the very large deformations... Would this improve if I move to msm_hocr?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/ecr05/MSM_HOCR_macOSX/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AP5rvC6U5k4krMcALingSWyesVHiGFE9ks5tI1T2gaJpZM4RXuRX .
r03ert0
Hello! Thank you for MSM : ) I'm working with the version that comes with FSL, and although it often works very nicely, sometimes there are regions that get veery large deformations. Is there a way of limiting the maximum amount of deformation allowed? I tried lowering down the lambdas, from 0,0.1,0.2,0.3 to 0,0.01,0.02,0.03, but I still get the very large deformations... Would this improve if I move to msm_hocr?