<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Hi Chao,</p>
    <p>I agree with you as far as my particular case is concerned.  But
      maybe it could be a good option for other people to have ?</p>
    <p>Kind regards,</p>
    <p>Vincent<br>
    </p>
    <div class="moz-cite-prefix">On 12.02.20 13:20, Chao Wu wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAH+sHrWrsw6hNiSepAosDOK_ndhDOTxajoUTDicgKORAni4TiQ@mail.gmail.com">
      <meta http-equiv="content-type" content="text/html; charset=UTF-8">
      <div dir="ltr">But streamerBP uses CPUOutputImageType so the 30Go
        is allocated on RAM instead of GRAM, so shouldn't be a
        problem...
        <div><br>
        </div>
        <div>Regards, Chao</div>
      </div>
      <br>
      <div class="gmail_quote">
        <div dir="ltr" class="gmail_attr">Simon Rit <<a
            href="mailto:simon.rit@creatis.insa-lyon.fr"
            moz-do-not-send="true">simon.rit@creatis.insa-lyon.fr</a>>
          于2020年2月12日周三 上午9:28写道:<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 dir="ltr">
            <div>Actually, the way I have implemented the streaming, it
              still allocates the 30Go complete volume and compute it
              piece by piece. One thing you could try is to remove the
              streamerBP object, connect directly the reconstruction to
              the writer
              "writer->SetInput(pfeldkamp->GetOutput());" and set
              the streaming in the writer
              "writer->SetNumberOfStreamDivisions(args_info.divisions_arg);".
              Then it never allocates the whole volume in memory. If
              that works for you, I think you can open a PR on github
              with this change, that makes a lot more sense in my
              opinion.</div>
          </div>
          <br>
          <div class="gmail_quote">
            <div dir="ltr" class="gmail_attr">On Tue, Feb 11, 2020 at
              8:46 PM vincent <<a href="mailto:vl@xris.eu"
                target="_blank" moz-do-not-send="true">vl@xris.eu</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 bgcolor="#FFFFFF">
                <p>Hi Simon,</p>
                <p>yes, I used both in my command line.  I have 64 Go
                  RAM on the machine, so that shouldn't be the issue. 
                  For the sake of completeness, I also tried the subset
                  option in combination with the divisions option, going
                  as low as 1, but to no avail.</p>
                <p>I'll investigate further tomorrow.</p>
                <p>Thank you again for your help,</p>
                <p>Vincent<br>
                </p>
                <div>On 2020-02-11 8:08 p.m., Simon Rit wrote:<br>
                </div>
                <blockquote type="cite">
                  <div dir="ltr">
                    <div>Have you tried the combination of both? To be
                      clear, --divisions acts on the reconstructed
                      volume so it should be ~7 Go with the "--divisions
                      4" option (instead of
                      2000*2000*2000*4/1024/1024/1024=29.8 Go
                      otherwise).</div>
                    <div>The --lowmem option acts on the projections and
                      you have 250 Mo (instead of
                      2048*2048*1500*4/1024/1024/1024=23.4 Go
                      otherwise).</div>
                    <div>The message "Failed to allocate memory for
                      image" seems to be a CPU memory issue. Are you
                      sure you have about 10 Go available to run this
                      reconstruction?<br>
                    </div>
                  </div>
                  <br>
                  <div class="gmail_quote">
                    <div dir="ltr" class="gmail_attr">On Tue, Feb 11,
                      2020 at 7:31 PM vincent <<a
                        href="mailto:vl@xris.eu" target="_blank"
                        moz-do-not-send="true">vl@xris.eu</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 bgcolor="#FFFFFF">
                        <p>Hi Simon, <br>
                        </p>
                        <p>I am afraid I forgot to mention something in
                          my last email.  I tried to use the lowmem
                          option, as you suggested a while ago in the
                          list for the same problem, but I am afraid I
                          am still getting the same error.</p>
                        <p>kind regards,</p>
                        <p>Vincent<br>
                        </p>
                        <div>On 11.02.20 17:36, Simon Rit wrote:<br>
                        </div>
                        <blockquote type="cite">
                          <div dir="ltr">
                            <div>Hi Vincent,</div>
                            <div>There is a way to do such a thing in
                              rtkfdk with the --divisions option, see
                              code <a
href="https://github.com/SimonRit/RTK/blob/master/applications/rtkfdk/rtkfdk.cxx#L190-L196"
                                target="_blank" moz-do-not-send="true">here</a>.
                              <br>
                            </div>
                            <div>I also don't really understand either
                              what's going on in your bottom
                              reconstruction, it seems to be a geometric
                              problem. Have you checked an axial slice?</div>
                            <div>Simon</div>
                          </div>
                          <br>
                          <div class="gmail_quote">
                            <div dir="ltr" class="gmail_attr">On Tue,
                              Feb 11, 2020 at 4:21 PM vincent <<a
                                href="mailto:vl@xris.eu" target="_blank"
                                moz-do-not-send="true">vl@xris.eu</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">Hello
                              RTK community,<br>
                              <br>
                              I am afraid that my question might not be
                              directly related to the <br>
                              excellent implementation we are all using,
                              but it might still be <br>
                              interesting for some of you.<br>
                              <br>
                              I have a stack of 1500 projections of size
                              2048*2048.  I obviously can't <br>
                              reconstruct the full resolution volume on
                              my graphics card, as it is too <br>
                              big.  So my solution was to split the
                              sinogram into N parts, for which <br>
                              each reconstructed volume would fit in my
                              GPU memory and then reassemble <br>
                              them.  I did a test with a 700*820*900
                              sinogram, that I cut in two parts <br>
                              of 700*410(+a small overlap)*900.<br>
                              <br>
                              While the reconstruction of the whole
                              volume was acceptable, I got a <br>
                              weird issue with the split ones: the one
                              corresponding to the top of the <br>
                              image is also ok, but the bottom one is
                              very blurry.  The three images <br>
                              can be found at the following links:<br>
                              <br>
                              <a href="https://ibb.co/vLk9ZhQ"
                                rel="noreferrer" target="_blank"
                                moz-do-not-send="true">https://ibb.co/vLk9ZhQ</a><br>
                              <a href="https://ibb.co/m4pm0LT"
                                rel="noreferrer" target="_blank"
                                moz-do-not-send="true">https://ibb.co/m4pm0LT</a><br>
                              <a href="https://ibb.co/Jyf1yKM"
                                rel="noreferrer" target="_blank"
                                moz-do-not-send="true">https://ibb.co/Jyf1yKM</a><br>
                              <br>
                              I used the same calibration parameters for
                              the three reconstruction.  I <br>
                              visually checked the split sinograms and
                              they looked fine.<br>
                              <br>
                              <br>
                              Any insight will be much appreciated !<br>
                              <br>
                              <br>
                              Thanks in advance,<br>
                              <br>
                              kindest regards,<br>
                              <br>
                              Vincent<br>
                              <br>
_______________________________________________<br>
                              Rtk-users mailing list<br>
                              <a
                                href="mailto:Rtk-users@public.kitware.com"
                                target="_blank" moz-do-not-send="true">Rtk-users@public.kitware.com</a><br>
                              <a
                                href="https://public.kitware.com/mailman/listinfo/rtk-users"
                                rel="noreferrer" target="_blank"
                                moz-do-not-send="true">https://public.kitware.com/mailman/listinfo/rtk-users</a><br>
                              <br>
                            </blockquote>
                          </div>
                        </blockquote>
                      </div>
                      _______________________________________________<br>
                      Rtk-users mailing list<br>
                      <a href="mailto:Rtk-users@public.kitware.com"
                        target="_blank" moz-do-not-send="true">Rtk-users@public.kitware.com</a><br>
                      <a
                        href="https://public.kitware.com/mailman/listinfo/rtk-users"
                        rel="noreferrer" target="_blank"
                        moz-do-not-send="true">https://public.kitware.com/mailman/listinfo/rtk-users</a><br>
                    </blockquote>
                  </div>
                </blockquote>
              </div>
            </blockquote>
          </div>
          _______________________________________________<br>
          Rtk-users mailing list<br>
          <a href="mailto:Rtk-users@public.kitware.com" target="_blank"
            moz-do-not-send="true">Rtk-users@public.kitware.com</a><br>
          <a
            href="https://public.kitware.com/mailman/listinfo/rtk-users"
            rel="noreferrer" target="_blank" moz-do-not-send="true">https://public.kitware.com/mailman/listinfo/rtk-users</a><br>
        </blockquote>
      </div>
    </blockquote>
  </body>
</html>