<div dir="ltr">hi emma<div><br></div><div>many answers can be found by reading through relevant tests. for example:</div><div><br></div><div><a href="https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Registration/RegistrationMethodsv4/test/itkBSplineSyNImageRegistrationTest.cxx">https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Registration/RegistrationMethodsv4/test/itkBSplineSyNImageRegistrationTest.cxx</a><br>
</div><div><br></div><div>these tests represent raw material for a v4 software guide which hans johnson is leading.</div><div><div><br></div><div>the fast variant of syn is similar to "the demons algorithm"</div>
<div><br></div><div>in that it is a specific algorithm, not a generic one. </div></div><div><br></div><div>so, to answer your questions:</div><div><br></div><div class="gmail_extra"><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div dir="ltr"><div>Is there a simple way to set the learning rate given a scaleEstimator for these methods (without using the ans registration helper ) or must one simply choose one at the beginning and hope for the best??</div>
</div></blockquote><div><br></div><div>yes - just set the learning rate at the beginning ( e.g 0.1 ). </div><div><br></div><div>syn takes care of rescaling over iterations internally. no scaleEstimator </div><div><br>
</div>
<div>is needed for displacement fields/diffeomorphisms because the parameters</div><div><br></div><div>all scale in the same way.</div><div><br></div><div>changes in x are equivalent to changes in y.</div><div><br></div>
<div>
changes in x are different than changes in theta.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">
<div><br></div><div>I am also finding it difficult to determine which optimizer these methods use as in the ANTS headers there are two options a GradientDescent and a conjugate gradient descent, any ideas??</div></div></blockquote>
<div><br></div><div>greedy syn uses its own optimization fast optimization of the diff objective. time varying syn uses gradient descent. memory concerns abound when you move beyond these first order approaches.</div><div>
</div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">
<div><br></div><div>I initially used the ITK libraries as they provided a method that I could begin using state of the art image processing algorithms without the steep learning curve, is there any pending documentation or software guide for the new ITKV4 framework.</div>
</div></blockquote><div><br></div><div>Great to hear! Keep at it.</div></div></div></div>