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

Re: Difficulties getting Enfuse Auto Align droplet to work

Expand Messages
  • Erik Krause
    ... Yes, of course. Fisheye mapping is not linear hence you can t simply shift the images. If you shoot a (checkerboard) tiled floor straight down with a
    Message 1 of 24 , Feb 26, 2008
    • 0 Attachment
      On Tuesday, February 26, 2008 at 19:00, Carel Struycken wrote:

      > I had similar issues but thought it was user (me) stupidity. BTW, are
      > there any special issues for alignment with fisheyes? Is there a
      > reason why PS CS3 auto-align would not work or would not work as well?

      Yes, of course. Fisheye mapping is not linear hence you can't simply
      shift the images.

      If you shoot a (checkerboard) tiled floor straight down with a normal
      (rectilinear) lens alle tiles are the same size. You can shift the
      image by one tile and all tiles will match. This is no possible with
      a fisheye, because the tiles are not only different in size but not
      rectangular, too.

      best regards
      Erik Krause
      http://www.erik-krause.de
    • Bruno Postle
      ... Thinking again, increasing the amount of downscaling should effectively increase the search area, so -s 2 ought to do exactly what you need. ... The
      Message 2 of 24 , Feb 26, 2008
      • 0 Attachment
        On Tue 26-Feb-2008 at 11:21 -0800, Milko K. Amorth wrote:

        >Thanks for the insight. I have tried to scale -s0 (in hope it would not
        >scale) and grid -g2 , but to no avail.
        >It stops working beyond a 20 pixel (buffer) play between the input
        >images. Good enough for now.

        Thinking again, increasing the amount of downscaling should
        effectively increase the search area, so '-s 2' ought to do exactly
        what you need.

        >Some options are not documented ie: -s, -g and -l or i could not find
        >documentation on it.

        The usage looks like this:

        -l Assume linear input files
        -s scale Scale down image by 2^scale (default: 1 [2x downsampling])
        -g gsize Break image into a rectangular grid (gsize x gsize) and attempt to find
        num control points in each section (default: 5 [5x5 grid] )

        (by linear input it means scene-referenced gamma=1 data)

        >I like your stereographic projections of your handhelds.

        Thanks!

        --
        Bruno
      • Milko K. Amorth
        Hi Erik, ... In case of the stacking it seems to work fairly well. I took the same input sources and imported them to layers, initiated an auto align layers
        Message 3 of 24 , Feb 26, 2008
        • 0 Attachment
          Hi Erik,
          >
          >
          > Yes, of course. Fisheye mapping is not linear hence you can't simply
          > shift the images.
          >
          In case of the stacking it seems to work fairly well. I took the same
          input sources and imported them to layers, initiated an auto align
          layers process and exported layers to files.
          Tokk thes files in enfuse and got a good result. Auto align shifted and
          rotated the layers.
          The comparison is updated to see the results at:
          http://360image.de/test/align_enfuse.htm

          Carel, it better work for that kind of money! :-)

          Cheers, Milko
        • Milko K. Amorth
          Hi Bruno, ... I have tried those values before and just tried it again and it did fail. The terminal confirms only 2 from 3 images are read in. This was the
          Message 4 of 24 , Feb 26, 2008
          • 0 Attachment
            Hi Bruno,
            >
            >
            > Thinking again, increasing the amount of downscaling should
            > effectively increase the search area, so '-s 2' ought to do exactly
            > what you need.
            >
            I have tried those values before and just tried it again and it did
            fail. The terminal confirms only 2 from 3 images are read in.
            This was the set within a 30 pixel canvas buffer. The set within the 20
            pixel buffer works well.

            Thanks for the brainstorming.

            Cheers, Milko
          • Yuval Levy
            ... I ve just built the latest versions. No, they will not help you on this specific issue, but it means that I could put some work on making changes. I have
            Message 5 of 24 , Feb 26, 2008
            • 0 Attachment
              Milko K. Amorth wrote:
              > I have tried ...

              I've just built the latest versions. No, they will not help you on this
              specific issue, but it means that I could put some work on making
              changes. I have looked into align_image_stack.

              Changing the hard coded value for local search area width is IMO no
              solution. It may be a quick fix for this specific case, but I already
              see a whole lot of specific cases coming. The real fix for this case is
              to simply run it through a fully fledged optimization process, i.e. the
              real Hugin.

              Before spending time on modifying align_image_stack, I would like to
              understand the utility of such exercise. Why expand what was meant to be
              a reduced, scope limited version; only to reproduce functionality that
              already exists in its bigger sibling, the fully fledged Hugin?

              Carlos, have you tried to drop all the image files on Hugin (rather than
              on the droplet) and see what comes out of it?

              If you are more comfortable with PTgui, you can drop them on PTgui,
              align them there, then import the PTgui project into hugin for fusing.

              If Hugin or PTgui fail to align the images, there is nothing that
              align_image_stack can do.

              Yuv
            • Milko K. Amorth
              Hi Yuv, ... I must agree with you. It does not make sense. The exercise of doing this comparison was for me to find its limits. As I have told you already, I
              Message 6 of 24 , Feb 27, 2008
              • 0 Attachment
                Hi Yuv,
                >
                >
                > Before spending time on modifying align_image_ stack, I would like to
                > understand the utility of such exercise. Why expand what was meant to be
                > a reduced, scope limited version; only to reproduce functionality that
                > already exists in its bigger sibling, the fully fledged Hugin?
                >
                I must agree with you. It does not make sense. The exercise of doing
                this comparison was for me to find its limits.
                As I have told you already, I am quite happy with what it does already.
                I just like to get to know the tools in my arsenal.
                I agree that align_image_stack has its place a module or standalone in a
                powerful stitching suite like Hugin and compliments it.
                Its not long that it has been introduced and I just swiped it when I
                first have read about it in Sept 2007.
                When Peters struggle with PTgui came up to try to stack these images for
                enfuse it was the first thing that came to my mind to try,
                since I had not before.

                Using Hugin or Ptgui or other Panorama Tools based Guis will do what
                Carlos wants it to do.
                It's just a matter of knowing how to use the tools you already have on
                hand. There are so many ways to achieve one result.
                That does not mean one is better than the other. Its just a matter of
                ability , preference and means.

                Thanks Yuv and Bruno, for looking into this and helping out.

                Cheers, Milko
              Your message has been successfully submitted and would be delivered to recipients shortly.