<div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div dir="ltr"><div>Dear Laura,</div><div>Interesting, this option is mainly here for improving the memory management. All it does is split your input stack in subsets to do the reconstruction, thus only loading a few projections in the CPU and GPU memory. This is not implemented in the backprojection because we usually keep the projections in memory during iterative reconstruction.</div><div>It could be implemented in C++ but I maybe you could not try do it in Python too: instead of calling backprojection once, split your input projection stack and geometries in pieces and call backprojection several times (except maybe for some memory transfers which cannot be avoided). Such a subselection is done here (in c++):</div><div><a href="https://github.com/SimonRit/RTK/blob/master/applications/rtksubselect/rtksubselect.cxx">https://github.com/SimonRit/RTK/blob/master/applications/rtksubselect/rtksubselect.cxx</a></div><div>You can open an issue if you feel it's worth implementing the option in the backprojection, it should be doable but is not straightforward.</div><div>Cheers,</div><div>Simon<br></div><div><br></div></div></div></div></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, Sep 9, 2022 at 10:22 AM Laura Hellwege <<a href="mailto:hellwege@imt.uni-luebeck.de" target="_blank">hellwege@imt.uni-luebeck.de</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div>
<div lang="DE">
<div>
<p class="MsoNormal"><span lang="EN-US">Dear all,<u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US"><u></u> <u></u></span></p>
<p class="MsoNormal"><span lang="EN-US">I am using the filter “rtkCudaBackProjectionImageFilter” with Python-Wrapping for a self-implemented iterative algorithm. Right now, this backprojection (and therefore my algorithm) is relatively slow since it does not
use the full capacity of my GPU.<u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US">I know that “rtkCudaFDKConeBeamReconstructionFilter” has the option “SetProjectionSubsetSize()” that regulates how many projections are simultaneously processed. Using this speeds up my FDK reconstructions significantly.
<u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US">So my question is: Is something similar already implemented for the simple backprojection?<u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US"><u></u> <u></u></span></p>
<p class="MsoNormal"><span lang="EN-US">Best regards, <u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US">Laura<u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US"><u></u> <u></u></span></p>
<p class="MsoNormal"><span lang="EN-US">PS: Unfortunately, I have no background in C++ programming. Hence, to add this option myself I probably have to rely on expertise and support from this community.<u></u><u></u></span></p>
<p class="MsoNormal"><span lang="EN-US"><u></u> <u></u></span></p>
</div>
</div>
_______________________________________________<br>
Rtk-users mailing list<br>
<a href="mailto:Rtk-users@public.kitware.com" target="_blank">Rtk-users@public.kitware.com</a><br>
<a href="https://public.kitware.com/cgi-bin/mailman/listinfo/rtk-users" rel="noreferrer" target="_blank">https://public.kitware.com/cgi-bin/mailman/listinfo/rtk-users</a><br>
</div></blockquote></div>