Loading ...
Sorry, an error occurred while loading the content.

Re: [PanoToolsNG] Re: Composite high resolution panorama alignment help needed

Expand Messages
  • Matthias Taugwalder
    Hi Bernhard ... Yes, that is the exact same behaviour I am experiencing. I may imagine that this problems could result from many medium bad control points
    Message 1 of 24 , Apr 29 10:12 PM
    • 0 Attachment
      Hi Bernhard

      > I don't use APClean any more as PTGui offers the same functionality with
      > CP cleaning and by sorting the control points by distance.
      > As i wrote to Matthias, the bad control points aren't really bad, just
      > "optimized wrong" and with my usual methods (putting weight on good
      > links), the optimizer didn't "pop" into a new optimum.

      Yes, that is the exact same behaviour I am experiencing. I may imagine
      that this problems could result from many "medium bad" control points
      that increase the average control point distance. And the really good
      controlpoints don't have enough influence anymore on the whole image,
      and get missaligned.

      It would be nice if PTgui could offer a kind of weighting of control
      points, so that for example you could priorize and influence the
      optimization process, that you can optimize for example for the
      "important images" at the horizon.

      Bernhard, you may want to try this: Just optimize for one single row
      of pictures (means one row with pictures of all four images), by
      selecting/deselecting the other images in the optimizer tab. Just
      optimize for the individual yaw/pitch/roll values, don't optimize
      globally. Then in a 2nd step only optimize for the 2nd row and so on.
      May be you can outflank the optimizer that way...


      Have a nice day,

      Matthias

      --
      Matthias Taugwalder
      matthias.taugwalder@...
    • Bernhard Vogl
      ... PTGui also offers CP cleaning by statistical differences, but it isn t that configurable like APClean. so, thanks Erik for pointing out that it is worth
      Message 2 of 24 , Apr 30 12:26 AM
      • 0 Attachment
        > The PTGui CP cleaning tool is by no means as flexible and effective as
        > APClean. When PTGui has removed all "worst controlpoints" and the
        > panorama still doesn't fit very well APClean can enhance the result in
        > most cases.

        PTGui also offers CP cleaning by statistical differences, but it isn't that configurable like APClean. so, thanks Erik for pointing out that it is worth trying!
        I have made 2 runs with APClean yesterday. It reminded me why i normally don't use it - it is a good software but re-loading a large panorama afterwards takes ages...
        The optimizer run afterwards did change some details but the main problem is still there. I slowly start to believe that my specific optimization problem has no geometric solution.

        Bernhard
      • Bernhard Vogl
        ... That s a common problem i m facing with large panoramas. In that case, i normally sort the CPs by distance. If the CPs are correct but the distance high, i
        Message 3 of 24 , Apr 30 12:33 AM
        • 0 Attachment
          > Yes, that is the exact same behaviour I am experiencing. I may imagine
          > that this problems could result from many "medium bad" control points
          > that increase the average control point distance. And the really good
          > controlpoints don't have enough influence anymore on the whole image,
          > and get missaligned.
          >
          > It would be nice if PTgui could offer a kind of weighting of control
          > points, so that for example you could priorize and influence the
          > optimization process, that you can optimize for example for the
          > "important images" at the horizon.

          That's a common problem i'm facing with large panoramas. In that case, i normally sort the CPs by distance. If the CPs are correct but the distance high, i put weight on that image pair by increasing the number of control points. After repeating this step several times, the panorama usually "pops" into a correct optimization.

          > Bernhard, you may want to try this: Just optimize for one single row
          > of pictures (means one row with pictures of all four images), by
          > selecting/deselecting the other images in the optimizer tab. Just
          > optimize for the individual yaw/pitch/roll values, don't optimize
          > globally. Then in a 2nd step only optimize for the 2nd row and so on.
          > May be you can outflank the optimizer that way...

          Phew - there are soo many rows... ;-)
          I now took your first advice and started tu build each section separately.
          I wonder into what type of horror the final assembly will lead but maybe it's actually easier to handle than 500 mis-aligned images :-)

          Bernhard
        • Matthias Taugwalder
          ... Just an idea: You might want to write a very small script that parses the pts files and generates the images selected in the template file. It should be
          Message 4 of 24 , Apr 30 12:42 AM
          • 0 Attachment
            > Phew - there are soo many rows... ;-)
            > I now took your first advice and started tu build each section separately.
            > I wonder into what type of horror the final assembly will lead but maybe
            > it's actually easier to handle than 500 mis-aligned images :-)

            Just an idea: You might want to write a very small script that parses
            the pts files and generates the images selected in the template file.
            It should be quite easy to do, that way you can also e.g. specify the
            row to select as an argument. So, call the script for row1, open the
            generated template, optimize, close, call the script again for row2,
            and so on...


            Matthias

            --
            Matthias Taugwalder
            matthias.taugwalder@...
          • pedro_silva58
            bernhard, not sure what you mean by re-loading a large panorama afterwards takes ages , but this may help: you don t have to quit ptgui to run apclean! what i
            Message 5 of 24 , Apr 30 12:52 AM
            • 0 Attachment
              bernhard,

              not sure what you mean by "re-loading a large panorama afterwards takes ages", but this may help: you don't have to quit ptgui to run apclean!
              what i sometimes do is
              1- with the project open in ptgui, save under a different name (just in case). keep the project open in ptgui.
              2- open apclean, run it on the new project.
              3- in ptgui, go file-recent_projects and re/load the same (open) project. since all temp files are the same, this is fast (at least on my projects, which i realize are small by some people's standards). reoptimize, etc.
              4- if needed, soap, wash, rinse, etc.

              hth,
              pedro

              --- In PanoToolsNG@yahoogroups.com, "Bernhard Vogl" <bvogl@...> wrote:
              ...
              > I have made 2 runs with APClean yesterday. It reminded me why i normally don't use it - it is a good software but re-loading a large panorama afterwards takes ages...
              ...
              > Bernhard
            • Jeffrey Martin
              bernhard, you should talk to Luca Vascon, he is the TOP WARLOCK in my opinion regarding this type of image. One reason which no one i think mentioned so far
              Message 6 of 24 , Apr 30 2:28 AM
              • 0 Attachment
                bernhard,

                you should talk to Luca Vascon, he is the TOP WARLOCK in my opinion
                regarding this type of image.

                One reason which no one i think mentioned so far that you might have some
                funny problems is that if you shoot a tower pano with many meters away from
                the nodal point, you are in fact capturing more than 360 degrees, and so if
                you do everything properly, you'll end up with a curved horizon.

                for example, this pano which is by no means perfect but is still one of my
                favorites :)
                http://www.360cities.net/image/venice-skyline-the-magic-view

                the final equirect is 7500x3750. Usually with this setup (5d + 10.5) the
                final equirect should be less than 7000x3500. I stitched 4 corners and then
                joined them manually (with a STRAIGHT horizon).

                There are a couple other steps in this process, Bernhard you're a smart guy
                and can figure it out, I regret to say that I signed a blood oath with Dr.
                Vascon so if anyone wants to learn this particular process, you have to go
                to Venice yourself and ask him ;-))

                Jeffrey


                [Non-text portions of this message have been removed]
              • Bernhard Vogl
                ... Oh my God - I ve missed that feature! Thanks Pedro!
                Message 7 of 24 , Apr 30 3:49 PM
                • 0 Attachment
                  > not sure what you mean by "re-loading a large panorama afterwards takes ages", but this may help: you don't have to quit ptgui to run apclean!
                  >
                  >
                  Oh my God - I've missed that feature!
                  Thanks Pedro!
                • Ken Warner
                  I ve noticed sometimes after adding some CP s manually after optimization. Those CP s can be seen by PTGui as being very far apart even though they are nearly
                  Message 8 of 24 , May 1, 2009
                  • 0 Attachment
                    I've noticed sometimes after adding some CP's manually after
                    optimization. Those CP's can be seen by PTGui as being very
                    far apart even though they are nearly perfectly placed. And
                    optimization doesn't seem to reduce the distance between those
                    newly added CP's.

                    I don't know what that means. Just thought I'd mention it in
                    this conversation.

                    Erik Krause wrote:
                    > Bernhard Vogl wrote:
                    >
                    >
                    >>>Did you try APClean to remove bad control points? It sometimes works
                    >>>wonders, but you need more than enough points of course.
                    >>>
                    >>
                    >>I don't use APClean any more as PTGui offers the same functionality with
                    >>CP cleaning and by sorting the control points by distance.
                    >
                    >
                    > The PTGui CP cleaning tool is by no means as flexible and effective as
                    > APClean. When PTGui has removed all "worst controlpoints" and the
                    > panorama still doesn't fit very well APClean can enhance the result in
                    > most cases.
                    >
                    > And deleting CP's by distance isn't a good solution either. If in an
                    > image pair most of the CP's have large errors they are probably
                    > optimized wrong not positioned wrong. APClean takes account of this
                    > since it uses the standard deviation as a measure.
                    >
                    > best regards
                  Your message has been successfully submitted and would be delivered to recipients shortly.