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

making .mov - better workflow for Hugin, Pano2QTVR and other

Expand Messages
  • dorindxn
    Hi list, I ve posted this on panoramic forum where I received only a single quality reply from John Houghton, as John usually does, others seems to not be
    Message 1 of 5 , Jun 24, 2007
    • 0 Attachment
      Hi list, I've posted this on panoramic forum where I received only a
      single quality reply from John Houghton, as John usually does, others
      seems to not be interested or didn't use Hugin (still the workflow is
      suitable for PTGui too) so here is the issue

      ---------8<---------------------------

      For some time I study the way to obtain a better quality for QT .mov
      file, as I've said in an old post, we must fight with anything that
      reduce quality and that is for example remapping and blending, the
      better interpolator/blender the small is the lost of quality.

      To be better understand let's recap what is the workflow to obtain a
      good quality equirectangular tiff image then the .mov

      1. conversion from RAW to TIFF (aplying CA correction .. )
      2. done in Hugin session - control points, cropp, etc also does more
      or less transpareant the next two steps
      3. remap in many intermediate tiffs according with their position on
      equirectangular -stitching engine and interpolators used
      4. blend those intermediate tiffs into final equirectangular -
      blender used and lots of memory and processor time
      5. done in Pano2QTVR conversion of big equirectangular in six
      intermediate cubic faces - - interpolators used
      6. done in Pano2QTVR produce .mov from those 6 cubic faces.

      To reduce the lost of quality I was thinking to jump from step 2
      directly to step 6 using some special, at this time bit complicated
      steps, say step BC1 and step BC2 like:

      BC1. remap in intermediate tiffs according with their position on
      cubic faces -stitching engine and interpolators used

      BC2. blend those intermediate tiffs into final cubic face - blender
      used and lesser memory and processor time

      The avanced users of Hugin can already guess/know how it is done
      using rectiliniar projection 90 degree for both FOV and HOV, and
      desired dimension for cubic faces, to switch to the next cubic face
      numerical transform is used in follow sequence:

      cube_0 no transform
      cube_1 -90 yaw transform
      cube_2 -90 yaw transform again
      cube_3 -90 yaw transform again
      cube_4 -90 yaw transform again then -90 pitch
      cube_5 +180 pitch transform


      The results are very encouraging and the best of this benefit the
      zenith and nadir (avoiding that stretch) but all the faces are
      improoved.

      There only one problem left, I don't have blend between each cubic
      faces you might notice where the cubic face ends, I already have a
      sollution or two for that but that will be some MC (more complicated)
      steps so maybe you can share ideas with me


      note: for producing of a.mov with cubic faces 5000x5000 (5000*pi =
      some equirectangular about 15700 pixels wide) the necessary of RAM
      memory is reduced from 2.5 GB to about 600-700 MB including all
      running application on my Win2K, time also was significantly reduced

      --------------8<--------------------------

      many thanks,
      Dorin
    • Bruno Postle
      ... This will work, the individual cube faces will have a slightly better quality because you are missing one interpolation step, and you avoid the nasty
      Message 2 of 5 , Jun 24, 2007
      • 0 Attachment
        On Sun 24-Jun-2007 at 10:18 -0000, dorindxn wrote:
        >
        >BC1. remap in intermediate tiffs according with their position on
        >cubic faces -stitching engine and interpolators used
        >
        >BC2. blend those intermediate tiffs into final cubic face - blender
        >used and lesser memory and processor time

        This will work, the individual cube faces will have a slightly
        better quality because you are missing one interpolation step, and
        you avoid the nasty artefacts blending produces in the poles of
        equirectangular images.

        ..but the cube-faces won't match, there is no way to persuade
        enblend to place seams consistently on every cubeface.

        I also think you are overstating the interpolation error of going
        via an intermediate equirectangular image. Normal images don't
        really lose much even with several interpolation steps - Unless they
        have been sharpened beforehand, but the solution here is to leave
        sharpening until last.

        --
        Bruno
      • dorindxn
        ... blender ... they ... Many thanks Bruno, usually I don t shapenning until last. First John s suggetion and aslo my thinking was to produce a 120 degrees
        Message 3 of 5 , Jun 25, 2007
        • 0 Attachment
          --- In PanoToolsNG@yahoogroups.com, Bruno Postle <bruno@...> wrote:
          >
          > On Sun 24-Jun-2007 at 10:18 -0000, dorindxn wrote:
          > >
          > >BC1. remap in intermediate tiffs according with their position on
          > >cubic faces -stitching engine and interpolators used
          > >
          > >BC2. blend those intermediate tiffs into final cubic face -
          blender
          > >used and lesser memory and processor time
          >
          > This will work, the individual cube faces will have a slightly
          > better quality because you are missing one interpolation step, and
          > you avoid the nasty artefacts blending produces in the poles of
          > equirectangular images.
          >
          > ..but the cube-faces won't match, there is no way to persuade
          > enblend to place seams consistently on every cubeface.
          >
          > I also think you are overstating the interpolation error of going
          > via an intermediate equirectangular image. Normal images don't
          > really lose much even with several interpolation steps - Unless
          they
          > have been sharpened beforehand, but the solution here is to leave
          > sharpening until last.
          >
          > --
          > Bruno
          >

          Many thanks Bruno, usually I don't shapenning until last.

          First John's suggetion and aslo my thinking was to produce a 120
          degrees then to cropp at 90 each cubic faces, in that way the
          adjacent areas next to seams where taking in consideration by enblend
          for each cubic face side, that prooved to leave minor but vissible
          seeams.

          At this point the result encourages me to write a blender to fix that
          after, for each 12 corner of the cube by process to match the pixels
          in a side area, that of course will not fix missing hands or so but
          for majority of cases will do the job.

          Another idea is to produce low resolution cubic (say 1500x1500) faces
          via intermediate equirectangular as usual, then via a mask to keep
          oversized to 5000x5000 only next to corner, that can be implemented
          and will act as kind of blur blender.

          any ideas are welcome, many thanks,

          Dorin
        • dorindxn
          Hi again, What if I produced 91x91 degrees faces then not to crop to 90, instead to overlap 4 cubic faces and to blend via enblend but how to tell to embleng
          Message 4 of 5 , Jun 25, 2007
          • 0 Attachment
            Hi again,

            What if I produced 91x91 degrees faces then not to crop to 90,
            instead to overlap 4 cubic faces and to blend via enblend but how to
            tell to embleng that, say, 4 faces of 5088x5088 must be blended in
            20000x5000 with -w parametter enabled that will asure over 180
            blending, the cubic faces to be at 0,5000,10000,15000 x offset? and
            by appling of only crop an rotation on cube_4 and cube_5 to complet
            that 20000x5000 on top and bottom for each with 4 portions from
            cube_4 and cube_5 and to blend to
            , say, 20000x5200.

            Note by applying only cropping and 90 degrees rotation for upper and
            lower "stripp" is no damage to image quality.

            So it's easy to wrote a software to prepare from those 6 faces
            5088x5088 those 12 pieces, as described above, to be blendend and to
            launch enblend then to reassamble them in 6 5000x5000 cubic faces but
            how to tel to enblend exact offset for each piece in final output?

            Maybe instead of 91 degress (5088x5088) I may use ,say, 90.3 or such


            Dorin
          • dorindxn
            Hi, finally the problem is solved, will consist in producing/croping at 91 degrees then manual blending with little overlap (with help of a little editor, yet
            Message 5 of 5 , Jun 25, 2007
            • 0 Attachment
              Hi, finally the problem is solved, will consist in producing/croping at
              91 degrees then manual blending with little overlap (with help of a
              little editor, yet to be wrote, that will show each cubic face with 4
              layers from adjacent faces, properly rotated so on.. ) due to
              recquirement that linear features must intersect exacly on the edge
              othwerwise they will feature as broken.

              thanks for help and reading,
              Dorin
            Your message has been successfully submitted and would be delivered to recipients shortly.