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

Re: Stitcher 5.5 Unlimited compared with PTGui 5.8 in auto-stitching mode

Expand Messages
  • robert_harshman
    Michel, I finally got a copy of Stitcher 5.5 Unlimited and have been doing some test myself in a series (Canon 10D with Nikor 10.5) that previously stitched
    Message 1 of 12 , Jul 31, 2006
    • 0 Attachment
      Michel,

      I finally got a copy of Stitcher 5.5 Unlimited and have been doing
      some test myself in a series (Canon 10D with Nikor 10.5) that
      previously stitched quite well with PTGui. My first couple of
      efforts with ST5.5U were less than stellar, more like unusable - too
      many stitching errors.

      Given what I've seen from your test, and now a few others using the
      Nikor 10.5 and ST5.5U I knew there must be some reason why my test
      were failing and yours and some others were working when using the
      10.5 on a non full frame camera.

      On your 20D and 10.5 test are you actually performing the "Calibrate
      Fisheye lens" or simply letting ST5.5U figure this out. That is,
      simply changing the Properties setting for Camera to Fisheye Full
      Frame, but NOT selecting the "Calibrate Fisheye lens" button. If I
      do that and select auto stitch I get a value of -343 for the
      distortion and an Angle of 166.950 after also using the Adjust
      Camera option after the auto stitch has completed.

      If I use the Calibrate Fisheye lens button and let it set the
      default - simply clicking OK to set the circle - I get quite
      different values for distortion -.284 and angle 164.706.

      What were your numbers?

      Also, all the successful stitches I've seen done with ST5.5U to date
      were of very tight spaces with lots of varying details. Perhaps this
      is whats needed to get a good template for bigger spaces?

      Any other ideas?

      Best Regards,

      Robert Harshman
    • michel thoby
      ... No, in all and every of my presented tests cases with a fisheye (and in many others that were not presented as well), I have selected the Calibrate
      Message 2 of 12 , Aug 1 1:10 AM
      • 0 Attachment
        Le 1 août 06 à 04:10, robert_harshman a écrit :

        > Given what I've seen from your test, and now a few others using the
        > Nikor 10.5 and ST5.5U I knew there must be some reason why my test
        > were failing and yours and some others were working when using the
        > 10.5 on a non full frame camera.
        >
        > On your 20D and 10.5 test are you actually performing the "Calibrate
        > Fisheye lens" or simply letting ST5.5U figure this out. That is,
        > simply changing the Properties setting for Camera to Fisheye Full
        > Frame, but NOT selecting the "Calibrate Fisheye lens" button.
        No, in all and every of my presented tests cases with a fisheye (and
        in many others that were not presented as well), I have selected the
        "Calibrate Fisheye Lens" button. Just to make sure and see by myself;)
        Completely differing from the 10,5 mm "Circular" case, I don't
        remember having had to adjust the circle from its default position
        (as it passes exactly by each of the corner of the rectangular
        image), though.

        > If I do that and select auto stitch I get a value of -343 for the
        > distortion and an Angle of 166.950 after also using the Adjust
        > Camera option after the auto stitch has completed.
        I believe that you meant -0.343?

        > If I use the Calibrate Fisheye lens button and let it set the
        > default - simply clicking OK to set the circle - I get quite
        > different values for distortion -.284 and angle 164.706.
        BTW: this is exactly what I did for the FF fisheye case in every
        occasion and as I described before.

        I had not noticed that such a difference could occur between the two
        ways: it looks like a bug to me.

        > What were your numbers?
        You shall find an answer by opening the .rzs files that are posted on
        my site (for all cases).
        The 10,5 mm/ 20D case is at: <http://michel.thoby.free.fr/
        PTGui5.8_&_ST5.5/Atelier%2010-5_20D/>
        One can read on the third line: fovx="163.58" and "rd="
        -2.38542e-008. The later number would be reported to the user as
        -0.238 in the "Properties" window.

        To be honest, I did not care to know exactly.
        1) My aim was to simulate a casual user workflow: after pushing the
        "AutoStitch" button and as soon I got an image, the next step has
        been to select "AutoAlign" and if not OK to select the manual
        alignment mode.
        Very soon after beginning my test campaign (i.e. before doing the
        presented article) I started not to read the parameters values
        anymore. Sorry.
        2) I had noticed very early during these training tests that Stitcher
        5.5 U computed lens parameters for the Nikkor 10,5 were obviously not
        accurate at all (w/respect to my own optical bench testing). I also
        have noted that a single parameter value for distortion description
        obviously could not accurately models the complex distortion shape of
        some fisheyes/sensor combinations.
        Nevertheless I have been surprised by the stitching quality that you
        could get even with this handicap;)

        > Also, all the successful stitches I've seen done with ST5.5U to date
        > were of very tight spaces with lots of varying details. Perhaps this
        > is whats needed to get a good template for bigger spaces?
        You may be right. I willingly have chosen this way leading to better
        automatic stitching rate and possibly associated with parallax errors
        rather than vice versa for these test runs.
        I have ran dozen of other tests, like you by using old projects from
        my hard disk without any failure to "stitch" (but two or three
        cases). I cannot tell you if there was not stitching errors then.

        > Any other ideas?
        Not much!

        Best,

        Michel
      • robert_harshman
        ... In the limited testing I ve done with circular fisheye stitching, ST5.5U has worked quite well, quick, easy perfect, stitches. At least using a Nikon 8400
        Message 3 of 12 , Aug 1 6:03 AM
        • 0 Attachment
          > "Calibrate Fisheye Lens" button. Just to make sure and see by
          > myself;) Completely differing from the 10,5 mm "Circular" case, I
          > don't remember having had to adjust the circle from its default
          > position (as it passes exactly by each of the corner of the
          > rectangular image), though.

          In the limited testing I've done with circular fisheye stitching,
          ST5.5U has worked quite well, quick, easy perfect, stitches. At
          least using a Nikon 8400 with an FE-8 adapter.

          The circle marking of the image area approach makes sense in my mind
          for circular fisheye lenses - i.e. where there is some black area
          showing the curve of the image circle. For full frame fisheye lenses
          I do not really understand how this could work as it seems to me
          that the "true" image circle is somewhere outside the corners of the
          image area, and very likely not at the corners. So marking the
          circle at the corners seems like it would not be right or the real
          edge of the image circle, just the edge of the frame. I guess
          perhaps the edge of the frame is all that matters, just does not
          seem logical to me. As the true edge of the image circle for a Nikor
          10.5 on a Canon 1.6 Camera must be quite a bit further out than the
          corners, otherwise this lens would not be a full frame on the Nikon
          1.5 Cameras it was design for.


          >
          > I believe that you meant -0.343?

          Yes, typing error.

          >
          > I had not noticed that such a difference could occur between the
          two ways: it looks like a bug to me.

          As a beta tester of 5.5 have you reported this to Realviz or should
          I?

          >
          > > What were your numbers?

          I guess I misunderstand these numbers and the use of them by the
          stitching process, as I would think that for a given camera/lens
          combo they should always be the same. And the numbers you get should
          be extremely close or even the same as the numbers I would get given
          the same basic setup.

          > Nevertheless I have been surprised by the stitching quality that
          you could get even with this handicap;)

          I've gone back and looked at my test with the 10.5 lens again and it
          seems to be the nadir and zenith shots are the cause of all
          stitching errors in these test. But, not one of the test was done in
          a small tight room with lots of varying details all around. I think
          I'll have to try shooting such a place to built a template.

          Michel, thank you!
        • John Houghton
          ... I m sure the important factor is that there is detail all around, not that it is a small tight room. In fact, you want as large an indoor venue as
          Message 4 of 12 , Aug 1 9:05 AM
          • 0 Attachment
            --- In PanoToolsNG@yahoogroups.com, "robert_harshman" <image360@...>
            wrote:
            > But, not one of the test was done in
            > a small tight room with lots of varying details all around. I think
            > I'll have to try shooting such a place to built a template.

            I'm sure the important factor is that there is detail all around, not
            that it is a small tight room. In fact, you want as large an indoor
            venue as possible - a church, maybe. Then there is less chance that
            parallax will interfere with the calibration.

            John
          • michel thoby
            ... John, I fully agree with you: I have myself looked around for a church in my property but I couln t find any;) Best regards, Michel
            Message 5 of 12 , Aug 2 1:40 AM
            • 0 Attachment
              Le 1 août 06 à 18:05, John Houghton a écrit :

              > --- In PanoToolsNG@yahoogroups.com, "robert_harshman" <image360@...>
              > wrote:
              >> But, not one of the test was done in
              >> a small tight room with lots of varying details all around. I think
              >> I'll have to try shooting such a place to built a template.
              >
              > I'm sure the important factor is that there is detail all around, not
              > that it is a small tight room. In fact, you want as large an indoor
              > venue as possible - a church, maybe. Then there is less chance that
              > parallax will interfere with the calibration.
              >
              > John

              John,

              I fully agree with you: I have myself looked around for a church in
              my property but I couln't find any;)

              Best regards,

              Michel
            • michel thoby
              ... I initially intended to do such a test reporting but I quickly decided not to pursue: There are so many important impacting factors to take into account
              Message 6 of 12 , Aug 2 1:58 AM
              • 0 Attachment
                Le 29 juil. 06 à 19:47, Mr. Roger Howard a écrit :

                >
                > On Jul 29, 2006, at 2:05 AM, michel thoby wrote:
                >
                >> Hi all,
                >>
                >> Stitcher Unlimited supports fisheye images. The first part of
                >> comparison tests with PTGui is reported here:
                >>
                >> http://michel.thoby.free.fr/PTGui5.8_&_ST5.5/SW_Compared.html
                >
                > Michel,
                >
                > Aside from cost and quality comparisons, the factor that interests me
                > most is in productivity. Is there any way you can report on two
                > aspects of this:
                >
                > 1) How much time did you spend, for each pano, setting up and editing
                > in the respective stitchers?
                > 2) How long does rendering take for each?
                >
                > Cheers,
                >
                > R

                I initially intended to do such a test reporting but I quickly
                decided not to pursue:
                There are so many important impacting factors to take into account
                and so much variation (understatement) from one beta -build number-
                version to the next and sometimes in a few hours span, that I soon
                got conflicting and not repeatable results. I gave up definitively on
                this matter.

                I would certainly recommend to make oneself opinion with its personal
                working environment and by testing the demos as soon as they are
                available.

                Regards,

                Michel
              • michel thoby
                Hi all, The second part of the comparison tests report is posted: http://michel.thoby.free.fr/PTGui5.8_&_ST5.5/SW_Compared_Part2.html Regards, Michel
                Message 7 of 12 , Aug 2 2:05 AM
                • 0 Attachment
                  Hi all,

                  The second part of the comparison tests report is posted:
                  http://michel.thoby.free.fr/PTGui5.8_&_ST5.5/SW_Compared_Part2.html

                  Regards,

                  Michel

                  Le 29 juil. 06 à 11:05, michel thoby a écrit :

                  > Hi all,
                  >
                  > Stitcher Unlimited supports fisheye images. The first part of
                  > comparison tests with PTGui is reported here:
                  >
                  > http://michel.thoby.free.fr/PTGui5.8_&_ST5.5/SW_Compared.html
                  >
                  > Regards,
                  >
                  > Michel
                • Oliver Mann
                  Hello Roger, as another beta tester of St5.5, I compared PTMac with the new Stitcher for speed and quality, using images of my S3 with 10.5 Nikkor. My first
                  Message 8 of 12 , Aug 2 10:29 AM
                  • 0 Attachment
                    Hello Roger,

                    as another beta tester of St5.5, I compared PTMac with the new
                    Stitcher for speed and quality, using images of my S3 with 10.5
                    Nikkor. My first results with automatic stitching led to fast
                    results, but the quality was not good enough (classic stitchin errors
                    at the seams). Since the high distortion tool for automatic lens
                    correction doesn´t work for fisheyes, I made a camera preset by hand.
                    I assembled a pano by manual stitching with control points. After
                    adjusting, I exported the camera preset. Now I can use this preset
                    for each new pano with the same lens, and the results are very close
                    to what I get with PTMac.
                    For rendering, I use "smart" - a new option which uses AFAIK enblend
                    algorithms. Currently this takes much too long. A 8000 x 4000px
                    rendering needs more than 2 hours on my G5. Doing the same with PTMac
                    and Enblend (or Stitcher Multi-TIFF output and Enblend standalone)
                    takes only 20 minutes, but at Realviz, they say that this will be
                    fixed in the next version. There is also an option to use an external
                    blending software, but there seems to be a bug, since I didn´t get it
                    to work.
                    In the current prerelease, there is still a lot of bugs. If Realviz
                    is able to fix them, this will be a very good software for the
                    professional workflow with fisheye images. Because of its WYSIWIG and
                    automatic stitching capabilities, preparing the images for stitching
                    is much faster as in PTMac. For good quality, a camera preset or
                    template is recommended. Both PTMac and Stitcher can create the same
                    quality, using enblend.

                    Oliver

                    Am 29.07.2006 um 19:47 schrieb Mr. Roger Howard:

                    > Aside from cost and quality comparisons, the factor that interests me
                    > most is in productivity. Is there any way you can report on two
                    > aspects of this:
                    >
                    > 1) How much time did you spend, for each pano, setting up and editing
                    > in the respective stitchers?
                    > 2) How long does rendering take for each?



                    [Non-text portions of this message have been removed]
                  • Roger Howard
                    ... Awesome, this is what I was looking to find out. Since quality was determined to be similar, or close enough, the only other factor that matters to me is
                    Message 9 of 12 , Aug 2 10:36 AM
                    • 0 Attachment
                      On Wed, August 2, 2006 10:29 am, Oliver Mann wrote:
                      > In the current prerelease, there is still a lot of bugs. If Realviz
                      > is able to fix them, this will be a very good software for the
                      > professional workflow with fisheye images. Because of its WYSIWIG and
                      > automatic stitching capabilities, preparing the images for stitching
                      > is much faster as in PTMac. For good quality, a camera preset or
                      > template is recommended. Both PTMac and Stitcher can create the same
                      > quality, using enblend.

                      Awesome, this is what I was looking to find out. Since quality was
                      determined to be similar, or close enough, the only other factor that
                      matters to me is performance (productivity), as frankly if it saves me
                      time over my existing process it will pay its steep price off quick enough
                      to be worth it.

                      I currently use PTGUI and, IMHO, that's already way more efficient than
                      PTMac (what I used to use) for setup of new panos... I can have a pano
                      ready to output in a few minutes at most, so Stitcher's GUI will be hard
                      pressed to beat that. But if the rendering itself is painfully slow then I
                      can't see it being worth the bother for now.

                      Perhaps for PTMac users it may be worth a new look when the performance
                      issues get solved?

                      What other features may draw people in? Anything really neat?
                    Your message has been successfully submitted and would be delivered to recipients shortly.