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

Stitching difficult panos in PTGui

Expand Messages
  • Keith Martin
    I use a 360Precision head for almost all my panos, and this means each shot is very precisely angled; stitching is never a problem. But occasionally I do shots
    Message 1 of 17 , Feb 25, 2008
    • 0 Attachment
      I use a 360Precision head for almost all my panos, and this means
      each shot is very precisely angled; stitching is never a problem.

      But occasionally I do shots in a rather less precise manner... for
      example, I got permission to take my camera into Heaven nightclub
      last week for the launch party for System 7's new album. (System 7 is
      Steve Hillage's band, a rather full-on trance/dance thing.) I took my
      Fuji with the 10.5mm lens, and I tried some over-the-head rotations
      on the offchance that I could stitch things together into *something*
      worth seeing.

      But I know that I've had little luck with PTGui and images taken that
      randomly, especially in a lively crowd with moving lights and
      projected images - entirely due to not really knowing how to control
      things properly, I'm sure. (I've also not had much luck stitching
      un-defished 10.5mm images in RealViz Stitcher.)

      Ermm... so does anyone have pointers?

      k
    • Rochy Liu
      hi all, what is the conversion factor in PTgui? I use NIKON E990,with 1536*2048 images,and focal length=7.7mm,but i do not know the conversion factor. who can
      Message 2 of 17 , Feb 25, 2008
      • 0 Attachment
        hi all,
        what is the conversion factor in PTgui? I use NIKON E990,with 1536*2048 images,and focal length=7.7mm,but i do not
        know the conversion factor.
        who can give me a hand,thx ....



        ____________________________________________________________________________________
        Never miss a thing. Make Yahoo your home page.
        http://www.yahoo.com/r/hs

        [Non-text portions of this message have been removed]
      • John Houghton
        ... Keith, You rightly aniticipate all sorts of stitching errors. I think the best thing to do is assign control points manually on the most distant, fixed
        Message 3 of 17 , Feb 25, 2008
        • 0 Attachment
          --- In PanoToolsNG@yahoogroups.com, Keith Martin <keith@...> wrote:
          >
          > Ermm... so does anyone have pointers?
          >
          Keith, You rightly aniticipate all sorts of stitching errors. I think
          the best thing to do is assign control points manually on the most
          distant, fixed features, which will be much less affected by parallax.
          Use calibrated lens parameters and don't optimize a,b & c. That should
          ensure reasonable integrity in the overall alignment. Smartblend will
          then hopefully produce a reasonable base image on which stitching
          errors can be tackled using layers and masks, copy/paste and cloning,
          transform tools, and anything else you can lay your hands on. Alpha
          channel masks in the input images may help too.

          John
        • AYRTON
          On Mon, Feb 25, 2008 at 11:55 AM, John Houghton ... Hi John, please why is that ? I mean not optmize the a,b,& c on these conditions
          Message 4 of 17 , Feb 25, 2008
          • 0 Attachment
            On Mon, Feb 25, 2008 at 11:55 AM, John Houghton <j.houghton@...>
            wrote:

            > --- In PanoToolsNG@yahoogroups.com, Keith Martin <keith@...> wrote:
            > >
            > > Ermm... so does anyone have pointers?
            > >
            > Keith, You rightly aniticipate all sorts of stitching errors. I think
            > the best thing to do is assign control points manually on the most
            > distant, fixed features, which will be much less affected by parallax.
            > Use calibrated lens parameters and don't optimize a,b & c.


            Hi John, please
            why is that ?
            I mean not optmize the a,b,& c on these conditions ???




            > That should
            > ensure reasonable integrity in the overall alignment. Smartblend will
            > then hopefully produce a reasonable base image on which stitching
            > errors can be tackled using layers and masks, copy/paste and cloning,
            > transform tools, and anything else you can lay your hands on. Alpha
            > channel masks in the input images may help too.


            To hide what exactly ??


            Thanks a lot

            AYRTON


            >
            >
            > John
            >
            >
            >
            > --
            > Yahoo! Groups Links
            >
            >
            >
            >


            --
            A Y R T O N
            + 55 21 9982 6313

            http://ayrton360.com
            http://rio.360cities.net
            http://vrfolio.com
            http://ayrton.com


            [Non-text portions of this message have been removed]
          • Hans Nyberg
            ... Well Smartblend is not useful for Keith as he is on Mac. But Photoshop autoblend is better than Smartblend for this. However If you have to much movements
            Message 5 of 17 , Feb 25, 2008
            • 0 Attachment
              --- In PanoToolsNG@yahoogroups.com, "John Houghton" <j.houghton@...> wrote:
              >
              > --- In PanoToolsNG@yahoogroups.com, Keith Martin <keith@> wrote:
              > >
              > > Ermm... so does anyone have pointers?
              > >
              > Keith, You rightly aniticipate all sorts of stitching errors. I think
              > the best thing to do is assign control points manually on the most
              > distant, fixed features, which will be much less affected by parallax.
              > Use calibrated lens parameters and don't optimize a,b & c. That should
              > ensure reasonable integrity in the overall alignment. Smartblend will
              > then hopefully produce a reasonable base image on which stitching
              > errors can be tackled using layers and masks, copy/paste and cloning,
              > transform tools, and anything else you can lay your hands on. Alpha
              > channel masks in the input images may help too.

              Well Smartblend is not useful for Keith as he is on Mac.

              But Photoshop autoblend is better than Smartblend for this.
              However If you have to much movements I guess the best would be to do it by hand in
              Photoshop.

              Do as John suggests use your standard values for the lens, but export it with standard
              blended and Layers. Than try the Autoblend on the Layers and see how that works.

              The rest is manual work.

              Hans
            • erik leeman
              ... Well, John advised to use pre-calibrated values for these parameters, optimizing them again using these messy images will most likely damage them. Using
              Message 6 of 17 , Feb 25, 2008
              • 0 Attachment
                --- In PanoToolsNG@yahoogroups.com, AYRTON <avi@...> wrote:
                > Hi John, please
                > why is that ?
                > I mean not optmize the a,b,& c on these conditions ???

                Well, John advised to use pre-calibrated values for these parameters,
                optimizing them again using these 'messy' images will most likely
                damage them. Using proven pre-calibrated values for your camera and
                lens gives you the best possible chance to obtain a usable stitch
                with a bad set of images.

                > >Alpha channel masks in the input images may help too.
                > To hide what exactly ??

                Carefully hiding nearby objects that you know will give the worst
                parallax problems on some of the shots can help prevent the most
                severe stitching errors.

                Regards,

                erik leeman

                (www.erikleeman.com)
              • Ken Warner
                Where does one get pre-calibrated values for a lens? How does one calibrate a lens? This is a very interesting topic and I d like to know a lot more about it
                Message 7 of 17 , Feb 25, 2008
                • 0 Attachment
                  Where does one get pre-calibrated values for a lens?

                  How does one calibrate a lens?

                  This is a very interesting topic and I'd like to know
                  a lot more about it but don't know where to begin looking.

                  Ken

                  erik leeman wrote:
                  > --- In PanoToolsNG@yahoogroups.com, AYRTON <avi@...> wrote:
                  >
                  >>Hi John, please
                  >>why is that ?
                  >>I mean not optmize the a,b,& c on these conditions ???
                  >
                  >
                  > Well, John advised to use pre-calibrated values for these parameters,
                  > optimizing them again using these 'messy' images will most likely
                  > damage them. Using proven pre-calibrated values for your camera and
                  > lens gives you the best possible chance to obtain a usable stitch
                  > with a bad set of images.
                  >
                  >
                  >>>Alpha channel masks in the input images may help too.
                  >>
                  >>To hide what exactly ??
                  >
                  >
                  > Carefully hiding nearby objects that you know will give the worst
                  > parallax problems on some of the shots can help prevent the most
                  > severe stitching errors.
                  >
                  > Regards,
                  >
                  > erik leeman
                  >
                  > (www.erikleeman.com)
                  >
                  >
                • Keith Martin
                  ... Where can I find some info on doing this properly? (Sorry, suddenly I feel like such a n00b!) I have actually managed to get a rough cylindrical stitch
                  Message 8 of 17 , Feb 25, 2008
                  • 0 Attachment
                    Sometime around 25/2/08 (at 15:56 +0000) erik leeman said:

                    >Well, John advised to use pre-calibrated values for these parameters,
                    >optimizing them again using these 'messy' images will most likely
                    >damage them. Using proven pre-calibrated values for your camera and
                    >lens gives you the best possible chance to obtain a usable stitch
                    >with a bad set of images.

                    Where can I find some info on doing this properly? (Sorry, suddenly I
                    feel like such a n00b!)

                    I have actually managed to get a rough cylindrical stitch done in
                    PTGui. I just added four or five control points for each image pair,
                    manually, then stitched as layered PSD. With some masking in
                    Photoshop I got this:

                    http://www.panoramapostcards.co.uk/vr/system7.html

                    The question now is can I take the stitched output as a flat TIFF and
                    add the zenith and nadir to it in PTGui? Or do I have to do it from
                    scratch when I add the top & tail?

                    k
                  • John Houghton
                    ... Hard to advise without knowing more about the images. Yes, you can input the edited image into PTGui (lens type: cylindrical, fov: 360), but it does
                    Message 9 of 17 , Feb 26, 2008
                    • 0 Attachment
                      --- In PanoToolsNG@yahoogroups.com, Keith Martin <keith@...> wrote:
                      >
                      > The question now is can I take the stitched output as a flat TIFF
                      > and add the zenith and nadir to it in PTGui? Or do I have to do it
                      > from scratch when I add the top & tail?

                      Hard to advise without knowing more about the images. Yes, you can
                      input the edited image into PTGui (lens type: cylindrical, fov: 360),
                      but it does appear to have particularly large holes to fill at the
                      zenith and nadir of roughly 90 degrees. In working with a
                      cylindrical output, you have lost a lot of vertical fov.

                      You could try starting a new project with the nadir and zenith
                      images. Add the stitched image, specifying individual parameters for
                      it (details are entered on the Image Parameters tab). Set control
                      points between the zenith and nadir images and the stitched image (if
                      possible) and optimize only the nadir and zenith to bring them into
                      alignment with the stitched image. Alternatively, manuipulate the
                      zenith and nadir in the Panorama Editor window to align them.
                      Depending how things pan out, you can stitch all three images
                      together to tiff, or output equirectangular layers and merge
                      manually, possibly generating cube faces to have better control. The
                      generally black background is very forgiving of stitching errors and
                      holes, which ought to help a lot.

                      It would probably be better to start again by editing an
                      equirectangular layered output, but even then, you might save time by
                      remapping your cylindrical stitch to equirectangular, and layer that
                      as well to merge in the work that you have already done.

                      John
                    • John Houghton
                      ... Ken, There is more than one way to do this, but one way is to just make a panorama as per normal, exercising sufficient care to get a really good result.
                      Message 10 of 17 , Feb 26, 2008
                      • 0 Attachment
                        --- In PanoToolsNG@yahoogroups.com, Ken Warner <kwarner@...> wrote:
                        >
                        > How does one calibrate a lens?

                        Ken, There is more than one way to do this, but one way is to just
                        make a panorama as per normal, exercising sufficient care to get a
                        really good result. This means avoiding things that normally might
                        compromise stitching accuracy. Ideally, choose a large cavernous
                        venue that has plentiful features for control points all around (and
                        above) but far enough away so that parallax isn't an issue. Site the
                        tripod on a firm footing and set up the head with the lens accurately
                        at the no-parallax point. Include zenith and nadir shots taken with
                        the camera on the pano head, and contrive a 50% overlap somewhere.

                        Assign control points carefully by hand so that they are well spread
                        out over the overlaps - not bunched in the middle, as often happens
                        with the auto generator. You don't need a huge number. Avoid
                        placing points on anything near the camera.

                        Set the image parameters to their known nominal values. Then
                        optimize, aiming to get the average control point distance to below 1
                        pixel and the maximum to less than 2. Check for stitching errors and
                        if found, check the control point placement and maybe assign some
                        extra points to those areas to coax the images into better alignment.

                        The resulting parameters can be copied into subsequent projects via
                        the template feature or (provided a crop hasn't been used) you can
                        save the lens parameters in the lens database.

                        John
                      • Keith Martin
                        ... Good point about the lack of info about the images! These are shots taken with a Fuji S5Pro and Nikkor 10.5mm lens, my normal pano setup, but hand-held
                        Message 11 of 17 , Feb 26, 2008
                        • 0 Attachment
                          > > The question now is can I take the stitched output as a flat TIFF
                          >> and add the zenith and nadir to it in PTGui? Or do I have to do it
                          >> from scratch when I add the top & tail?

                          >Hard to advise without knowing more about the images. Yes, you can
                          >input the edited image into PTGui (lens type: cylindrical, fov: 360),
                          >but it does appear to have particularly large holes to fill

                          :-)
                          Good point about the lack of info about the images!

                          These are shots taken with a Fuji S5Pro and Nikkor 10.5mm lens, my
                          normal pano setup, but hand-held over my head. The zenith image I
                          shot should be ample, the space should be no larger than if I'd done
                          it with my pano head and normal workflow.

                          The same goes for the nadir, although I'm sure it'll be rather more
                          of a manual hack-it-in job than normal: no details other than people,
                          and they weren't exactly standing still! Good job it is so dark down
                          there...

                          Is the output that I made there actually different in the (virtual)
                          height it covers because it wasn't an equirectangular output? That
                          seems odd, but I'm more than willing to consider that possibility if
                          it is one.

                          Thanks for the feedback and suggestions. I'll try importing the
                          stitched image and adding the zenith and nadir to that. If that
                          proves troublesome I'll go back to the PTGui project with the first
                          set of originals, and so on.

                          k
                        • John Houghton
                          ... Keith, It s hard to know what the vertical field of view of your panorama actually is. If you specified a cylindrical output, then you certainly won t
                          Message 12 of 17 , Feb 26, 2008
                          • 0 Attachment
                            --- In PanoToolsNG@yahoogroups.com, Keith Martin <keith@...> wrote:
                            >
                            > Is the output that I made there actually different in the (virtual)
                            > height it covers because it wasn't an equirectangular output?

                            Keith, It's hard to know what the vertical field of view of your
                            panorama actually is. If you specified a cylindrical output, then you
                            certainly won't have a full 180 degrees vertical coverage. With the
                            camera in portrait orientation, you should have 130 degrees or so.
                            DevalVR reports the view as slightly more than 90. I tried a
                            conversion of the .mov file with Pano2VR and got a 6647x3323 tiff -
                            supposedly equirectangular, though not exactly in the aspect ratio 2:1
                            because the width is an odd number. The visible image certainly seems
                            to be only around 90 degrees high.

                            John
                          • John Houghton
                            ... Keith, It s hard to know what the vertical field of view of your panorama actually is. If you specified a cylindrical output, then you certainly won t
                            Message 13 of 17 , Feb 26, 2008
                            • 0 Attachment
                              --- In PanoToolsNG@yahoogroups.com, Keith Martin <keith@...> wrote:
                              >
                              > Is the output that I made there actually different in the (virtual)
                              > height it covers because it wasn't an equirectangular output?

                              Keith, It's hard to know what the vertical field of view of your
                              panorama actually is. If you specified a cylindrical output, then you
                              certainly won't have a full 180 degrees vertical coverage. With the
                              camera in portrait orientation, you should have 130 degrees or so.
                              DevalVR reports the view as slightly more than 90. I tried a
                              conversion of the .mov file with Pano2VR and got a 6647x3323 tiff -
                              supposedly equirectangular, though not exactly in the aspect ratio 2:1
                              because the width is an odd number. The visible image certainly seems
                              to be only around 90 degrees high.

                              John
                            • Ken Warner
                              Hi John, I understand. But just for discussion -- this method could result in a slightly subjective result couldn t it? And it would depend on the correctness
                              Message 14 of 17 , Feb 26, 2008
                              • 0 Attachment
                                Hi John,

                                I understand. But just for discussion -- this method could
                                result in a slightly subjective result couldn't it?
                                And it would depend on the correctness of the software and
                                alignment of the hardware used to view the result. Or do
                                such fine adjustment really matter?

                                I've shot three or four cylindrical panos with a $10 USD
                                .45X wide angle lens adapter and they look ok to me even
                                though they probably are not. I suppose a spherical
                                pano would be more telling.

                                The point is that it's hard for me to tell
                                what is distorted and what is not unless the distortions
                                are gross. The stitchers do such a good job of straightening
                                things out that most panos look fine to me.

                                What if one used an image of a specific pattern and then
                                wrote some kind of image analysis program to detect the
                                difference between the actual image and expected image?

                                Or perhaps an interactive program where one straightened
                                lines and circles etc. by hand to conform to a nominal
                                image and then those corrections could be translated into a set
                                of calibration parameters. Writing an interactive program
                                would be a little easier writing than an automatic program.

                                Have you ever heard of such a program? This idea just popped
                                into my head so I haven't done an internet search yet. I'll
                                look around and if I find something I'll report it.

                                Ken


                                John Houghton wrote:
                                > --- In PanoToolsNG@yahoogroups.com, Ken Warner <kwarner@...> wrote:
                                >
                                >>How does one calibrate a lens?
                                >
                                >
                                > Ken, There is more than one way to do this, but one way is to just
                                > make a panorama as per normal, exercising sufficient care to get a
                                > really good result. This means avoiding things that normally might
                                > compromise stitching accuracy. Ideally, choose a large cavernous
                                > venue that has plentiful features for control points all around (and
                                > above) but far enough away so that parallax isn't an issue. Site the
                                > tripod on a firm footing and set up the head with the lens accurately
                                > at the no-parallax point. Include zenith and nadir shots taken with
                                > the camera on the pano head, and contrive a 50% overlap somewhere.
                                >
                                > Assign control points carefully by hand so that they are well spread
                                > out over the overlaps - not bunched in the middle, as often happens
                                > with the auto generator. You don't need a huge number. Avoid
                                > placing points on anything near the camera.
                                >
                                > Set the image parameters to their known nominal values. Then
                                > optimize, aiming to get the average control point distance to below 1
                                > pixel and the maximum to less than 2. Check for stitching errors and
                                > if found, check the control point placement and maybe assign some
                                > extra points to those areas to coax the images into better alignment.
                                >
                                > The resulting parameters can be copied into subsequent projects via
                                > the template feature or (provided a crop hasn't been used) you can
                                > save the lens parameters in the lens database.
                                >
                                > John
                                >
                                >
                              • John Houghton
                                ... It s true that different people will get slightly different results for a variety of reasons, but all may be satisfactory from a practical point of view.
                                Message 15 of 17 , Feb 26, 2008
                                • 0 Attachment
                                  --- In PanoToolsNG@yahoogroups.com, Ken Warner <kwarner@...> wrote:
                                  >
                                  > I understand. But just for discussion -- this method could
                                  > result in a slightly subjective result couldn't it?

                                  It's true that different people will get slightly different results
                                  for a variety of reasons, but all may be satisfactory from a
                                  practical point of view.

                                  > And it would depend on the correctness of the software and
                                  > alignment of the hardware used to view the result. Or do
                                  > such fine adjustment really matter?

                                  Nothing is perfect. Some stitchers are very much better than others,
                                  and some viewers are also better than others. Fine adjustment only
                                  matters if it produces a visibly better result. People generally
                                  find a workflow that gives them the quality of work that they need at
                                  an acceptable cost (time or money).

                                  > I've shot three or four cylindrical panos with a $10 USD
                                  > .45X wide angle lens adapter and they look ok to me even
                                  > though they probably are not. I suppose a spherical
                                  > pano would be more telling.

                                  It is considerably more telling, though a lot depends on how picky
                                  you are.


                                  > What if one used an image of a specific pattern and then
                                  > wrote some kind of image analysis program to detect the
                                  > difference between the actual image and expected image?

                                  That would be ideal and that sort of test is indeed done.
                                  Constructing a test image for a 180 degree lens is not
                                  straightforward, though. You may have seen my recent post which
                                  described a method of using the night sky as a test image:

                                  http://tech.groups.yahoo.com/group/PanoToolsNG/message/17962

                                  I used the optimizer to evaluate differences in the recorded
                                  positions of stars, but it would be much better to have a program
                                  that calculated the differences accurately.

                                  So many things conspire to produce minor imperfections when shooting
                                  day-to-day routine panoramas that it hardly seems worth spending a
                                  lot of time on attempting to get a perfect lens calibration, even if
                                  that were possible. However, I do think that spending a couple of
                                  hours calibrating a lens is time well spent.

                                  John
                                • Ken Warner
                                  Hi John, Very reasonable. The old saying, ...don t let perfect stand in the way of good... comes to mind. I saw the post on using stars etc. That seems
                                  Message 16 of 17 , Feb 26, 2008
                                  • 0 Attachment
                                    Hi John,

                                    Very reasonable. The old saying, "...don't let perfect stand in the way
                                    of good..." comes to mind.

                                    I saw the post on using stars etc. That seems like a pretty good way
                                    although it's harder than you might think to get a clear view of the
                                    sky these days -- depending on where you live. I doubt the people
                                    living in the SoCal Megalopolis have ever even seen stars.

                                    Anyway it's an interesting subject...

                                    Ken

                                    John Houghton wrote:
                                    > --- In PanoToolsNG@yahoogroups.com, Ken Warner <kwarner@...> wrote:
                                    >
                                    >>I understand. But just for discussion -- this method could
                                    >>result in a slightly subjective result couldn't it?
                                    >
                                    >
                                    > It's true that different people will get slightly different results
                                    > for a variety of reasons, but all may be satisfactory from a
                                    > practical point of view.
                                    >
                                    >
                                    >>And it would depend on the correctness of the software and
                                    >>alignment of the hardware used to view the result. Or do
                                    >>such fine adjustment really matter?
                                    >
                                    >
                                    > Nothing is perfect. Some stitchers are very much better than others,
                                    > and some viewers are also better than others. Fine adjustment only
                                    > matters if it produces a visibly better result. People generally
                                    > find a workflow that gives them the quality of work that they need at
                                    > an acceptable cost (time or money).
                                    >
                                    >
                                    >>I've shot three or four cylindrical panos with a $10 USD
                                    >>.45X wide angle lens adapter and they look ok to me even
                                    >>though they probably are not. I suppose a spherical
                                    >>pano would be more telling.
                                    >
                                    >
                                    > It is considerably more telling, though a lot depends on how picky
                                    > you are.
                                    >
                                    >
                                    >
                                    >>What if one used an image of a specific pattern and then
                                    >>wrote some kind of image analysis program to detect the
                                    >>difference between the actual image and expected image?
                                    >
                                    >
                                    > That would be ideal and that sort of test is indeed done.
                                    > Constructing a test image for a 180 degree lens is not
                                    > straightforward, though. You may have seen my recent post which
                                    > described a method of using the night sky as a test image:
                                    >
                                    > http://tech.groups.yahoo.com/group/PanoToolsNG/message/17962
                                    >
                                    > I used the optimizer to evaluate differences in the recorded
                                    > positions of stars, but it would be much better to have a program
                                    > that calculated the differences accurately.
                                    >
                                    > So many things conspire to produce minor imperfections when shooting
                                    > day-to-day routine panoramas that it hardly seems worth spending a
                                    > lot of time on attempting to get a perfect lens calibration, even if
                                    > that were possible. However, I do think that spending a couple of
                                    > hours calibrating a lens is time well spent.
                                    >
                                    > John
                                    >
                                    >
                                  • johncharlesriley
                                    ... Bet you are thinking of: Better is the enemy of good. Voltaire I try to remind myself of that when I get too perfectionistic on a project and just need
                                    Message 17 of 17 , Feb 28, 2008
                                    • 0 Attachment
                                      --- In PanoToolsNG@yahoogroups.com, Ken Warner <kwarner@...> wrote:
                                      >
                                      > Hi John,
                                      >
                                      > Very reasonable. The old saying, "...don't let perfect stand in the way
                                      > of good..." comes to mind.

                                      Bet you are thinking of:

                                      "Better is the enemy of good."
                                      Voltaire

                                      I try to remind myself of that when I get too perfectionistic on a project and just need to GET
                                      IT DONE!
                                    Your message has been successfully submitted and would be delivered to recipients shortly.