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

RE: FW: [PanoToolsNG] calibrating a lens/teleconverter combo using PTGui

Expand Messages
  • Serge Maandag
    ... Line type control points are processed as set. The way they work is as follows: You can define a line (within a panotools script) by starting it with t3,
    Message 1 of 13 , Jan 1, 2008
    • 0 Attachment
      > This leads me to the question - are straight line CPs processed in pairs
      > (per the CP numbers that show up in the images)? Or - are they processed in
      > sets (with all CPs that have the same line number/name in the 'Type' field
      > on the table below the images members of a single set)? Is there a rule of
      > thumb regarding where to designate the 2nd CP of each pair (clearly, very
      > close causes problems - that's what caused the optimizer to say 'too good to
      > be true').

      Line type control points are processed as set. The way they work is as follows:

      You can define a line (within a panotools script) by starting it with t3, t4 or any
      t followed by a number > 2. A line consists of 2 or more line segments. A line
      segment is defined by 2 coordinates, start and end of the segment.
      The optimizer will try to have all segments line up on one straight line.

      The way Joost added line type control points to PTGui might be a bit confusing.
      In most cases you just define 2 segments and want them to line up. In that case the
      logical way would be to click twice in the left pane (and thus define segment1) and
      then click twice in the right pane (defining segment2).

      Instead, Joost chose to have the user alternately click the left and right panes.
      You are doing exactly the same thing as above and in the same order, but you just
      keep switching panes. This means you can define more than 2 segments to define a
      line. This can be handy for long overhead wires. Just select the appropriate line in
      the "CP type" dropdown box before clicking left and right.

      To get the best results, always try to define your segments as big as possible. Be
      cautious though: When optimizing with line type control points, the optimizer aims
      to have the segments line up in the output image. Depending on the type of output
      projection, that may be valid or not at all!

      With rectilinear (flat) projection, all straight lines in real life are straight in
      the image, so it will be valid. With cylindrical or equirectangular projection it's
      only valid for vertical lines and for the horizon. If the line segments are close to
      each other it may still work or it may help to temporarily switch the output
      projection to rectilinear during the optimization.

      Serge.
      http://360photo.org/
    • Yoram Bernet
      Serge, thank you and thanks also to all the others who have been educating me on this. So ,Serge, this is really helpful, but it does change my understanding
      Message 2 of 13 , Jan 1, 2008
      • 0 Attachment
        Serge, thank you and thanks also to all the others who have been educating
        me on this. So ,Serge, this is really helpful, but it does change my
        understanding of how this works. Let me try.



        Let's say that I have a horizontal line in the image that is significantly
        curved due to barrel distortion. I had originally, been designating point
        pairs by clicking at opposite ends of the line (for the first pair of a
        line), then at various random midpoints for subsequent pairs (thinking that
        the algorithm tries to minimize the distance of all points from the line
        designated by the first and second CPs).



        Now based on what you're saying, it seems that I should do this differently.
        Specifically - it seems that I should designate the first point at one end
        of the line (say the left) and the second point at a point (probably roughly
        midway between the two ends of the image) at which the line is maximally
        distorted. That does it for the first pair. For the second pair, I should
        designate a point at the opposite far end of the line (the right) and the
        second point of the pair near to the second point of the first pair (the
        point of maximum distortion). - Is this correct? Wouldn't I then want to
        proceed by designating lots of subsegments along the line that I eventually
        want rendered straight?



        (Strictly speaking, I guess that what I'm calling the point of max
        distortion is actually the point of least distortion and the points near the
        edge of the image are the points of max distortion. )



        Another question - once I correct for barrel distortion, using this method,
        the nextproblem that shows up is perspective distortion (even though I've
        used a PC lens, there is some perspective distortion due to not having the
        sensor plane exactly parallel to the subject). To fix this, I'd like to be
        able to designate certain line segs and say "these should all be horizontal"
        and other segs and say "these should all be vertical". Is this done using
        the horizontal and vertical line types in the dialog box (t1 and t2)? Can I
        correct for both barrel distortion and perspective distortion in the same
        step or do you recommend doing so iteratively?



        Many thanks and Happy New Year!



        Y



        -----Original Message-----
        From: PanoToolsNG@yahoogroups.com [mailto:PanoToolsNG@yahoogroups.com] On
        Behalf Of Serge Maandag
        Sent: Tuesday, January 01, 2008 12:29 PM
        To: PanoToolsNG@yahoogroups.com
        Subject: RE: FW: [PanoToolsNG] calibrating a lens/teleconverter combo using
        PTGui



        > This leads me to the question - are straight line CPs processed in pairs
        > (per the CP numbers that show up in the images)? Or - are they processed
        in
        > sets (with all CPs that have the same line number/name in the 'Type' field
        > on the table below the images members of a single set)? Is there a rule of
        > thumb regarding where to designate the 2nd CP of each pair (clearly, very
        > close causes problems - that's what caused the optimizer to say 'too good
        to
        > be true').

        Line type control points are processed as set. The way they work is as
        follows:

        You can define a line (within a panotools script) by starting it with t3, t4
        or any
        t followed by a number > 2. A line consists of 2 or more line segments. A
        line
        segment is defined by 2 coordinates, start and end of the segment.
        The optimizer will try to have all segments line up on one straight line.

        The way Joost added line type control points to PTGui might be a bit
        confusing.
        In most cases you just define 2 segments and want them to line up. In that
        case the
        logical way would be to click twice in the left pane (and thus define
        segment1) and
        then click twice in the right pane (defining segment2).

        Instead, Joost chose to have the user alternately click the left and right
        panes.
        You are doing exactly the same thing as above and in the same order, but you
        just
        keep switching panes. This means you can define more than 2 segments to
        define a
        line. This can be handy for long overhead wires. Just select the appropriate
        line in
        the "CP type" dropdown box before clicking left and right.

        To get the best results, always try to define your segments as big as
        possible. Be
        cautious though: When optimizing with line type control points, the
        optimizer aims
        to have the segments line up in the output image. Depending on the type of
        output
        projection, that may be valid or not at all!

        With rectilinear (flat) projection, all straight lines in real life are
        straight in
        the image, so it will be valid. With cylindrical or equirectangular
        projection it's
        only valid for vertical lines and for the horizon. If the line segments are
        close to
        each other it may still work or it may help to temporarily switch the output
        projection to rectilinear during the optimization.

        Serge.
        http://360photo. <http://360photo.org/> org/





        [Non-text portions of this message have been removed]
      • Serge Maandag
        ... Yes, totally. ... No, The three coefficients a, b and that the panotools optimizer uses can only correct barrel and pincushion distortion. The two line
        Message 3 of 13 , Jan 1, 2008
        • 0 Attachment
          > Serge, thank you and thanks also to all the others who have been educating
          > Now based on what you're saying, it seems that I should do this differently.
          > Specifically - it seems that I should designate the first point at one end
          > of the line (say the left) and the second point at a point (probably roughly
          > midway between the two ends of the image) at which the line is maximally
          > distorted. That does it for the first pair. For the second pair, I should
          > designate a point at the opposite far end of the line (the right) and the
          > second point of the pair near to the second point of the first pair (the
          > point of maximum distortion). - Is this correct?

          Yes, totally.

          > Wouldn't I then want to
          > proceed by designating lots of subsegments along the line that I eventually
          > want rendered straight?

          No, The three coefficients a, b and that the panotools optimizer uses can only
          correct barrel and pincushion distortion. The two line segments are enough info for
          the optimizer to do that. Subsegments would only be useful to fix a wobbly line. But
          that wouldn't be possible anyway with only a,b and c.

          The less segments you define, the less confusing it is to the optimizer. If I have
          to correct a barrel shaped box, I define a line (consisting of 2 line segments) on
          each side of the box.

          > Another question - once I correct for barrel distortion, using this method,
          > the nextproblem that shows up is perspective distortion (even though I've
          > used a PC lens, there is some perspective distortion due to not having the
          > sensor plane exactly parallel to the subject). To fix this, I'd like to be
          > able to designate certain line segs and say "these should all be horizontal"
          > and other segs and say "these should all be vertical". Is this done using
          > the horizontal and vertical line types in the dialog box (t1 and t2)? Can I
          > correct for both barrel distortion and perspective distortion in the same
          > step or do you recommend doing so iteratively?

          Correct and yes, the optimizer can do it in one go. There are occasions where the
          optimizer goes haywire, but that rarely happens. If it does, you'll notice. Then you
          can always:

          - set t1/t2 points
          - optimize yaw pitch and roll
          - remove the the t1/t2 points and set line type points
          - turn off yaw pitch and roll optimization
          - optimize a, b and c (or at least b)

          Happy newyear to you to!

          Serge.
          http://360photo.org/
        • Erik Krause
          ... Sorry, this is wrong. The use of three coefficients (a, b and c) was especially necessary to allow the correction of wavy (or wobbly) distortion as well.
          Message 4 of 13 , Jan 4, 2008
          • 0 Attachment
            Serge Maandag-3 wrote:
            >
            >> Wouldn't I then want to
            >> proceed by designating lots of subsegments along the line that I
            >> eventually
            >> want rendered straight?
            >
            > No, The three coefficients a, b and that the panotools optimizer uses can
            > only
            > correct barrel and pincushion distortion. The two line segments are enough
            > info for
            > the optimizer to do that. Subsegments would only be useful to fix a wobbly
            > line. But
            > that wouldn't be possible anyway with only a,b and c.
            >
            > The less segments you define, the less confusing it is to the optimizer.
            > If I have
            > to correct a barrel shaped box, I define a line (consisting of 2 line
            > segments) on
            > each side of the box.
            >

            Sorry, this is wrong. The use of three coefficients (a, b and c) was
            especially necessary to allow the correction of wavy (or wobbly) distortion
            as well. Wavy distortion is a mixture of barrel and pincushion distortion in
            different distances to the image center. For details see
            http://wiki.panotools.org/Wavy_distortion

            To correct for this kind of distortion you need more than two pairs of
            control points on each line and you need lines in different distances from
            the center. The optimizer will not be confused at all, since there isn't a
            possibility of false local minima.

            best regards


            -----
            Erik Krause
            http://www.erik-krause.de
            --
            View this message in context: http://www.nabble.com/RE%3A-FW%3A-calibrating-a-lens-teleconverter-combo-using--PTGui-tp14569835p14615206.html
            Sent from the PanoToolsNG mailing list archive at Nabble.com.
          • Serge Maandag
            ... Thanks for the update, Erik. I kind of always pictured pincushion distortion as bulgy in the middle and pointy in the corners, but it occurs to me that
            Message 5 of 13 , Jan 4, 2008
            • 0 Attachment
              > Sorry, this is wrong. The use of three coefficients (a, b and c) was
              > especially necessary to allow the correction of wavy (or wobbly) distortion
              > as well. Wavy distortion is a mixture of barrel and pincushion distortion in
              > different distances to the image center. For details see
              > http://wiki.panotools.org/Wavy_distortion

              Thanks for the update, Erik.

              I kind of always pictured pincushion distortion as bulgy in the middle and pointy in
              the corners, but it occurs to me that this is in fact a combination of barrel
              distortion and pincushion distortion. More than this can't be correct by the
              optimizer, though.

              Luckily pincushion distortion is more rare than barrel distorion. Especially on wide
              angle lenses.

              Serge.
            • Yoram Bernet
              Folks: I was struggling with using the straight line control points to remove barrel distortion from a single image. I posted some questions to this list and
              Message 6 of 13 , Jan 24, 2008
              • 0 Attachment
                Folks:



                I was struggling with using the straight line control points to remove
                barrel distortion from a single image. I posted some questions to this list
                and got lots of helpful answers. Also - Erik - thanks for updating the
                tutorial.



                However, i'm still not getting the results that I hoped for. I shot a close
                up of a model using the Canon 24 mm tilt/shift lens. I used vertical shift.
                I'm trying to correct for barrel distortion.



                Please see the following illustration for reference
                www.yorambernet.com/Panotools/panotools.jpg. There are three images in the
                illustration: on the top left is my uncorrected image. On the top right is
                the corrected image and on the bottom, a snapshot of the PTGui 'Control
                Points' screen, showing the control points (CPs) that I used. I set CP pairs
                on three important lines and grouped them in the CP dialog box as new lines
                3, 4 and 5. On the top two images (corrected and uncorrected, I've inserted
                four light blue reference lines that are perfectly straight. The lower three
                of these correspond to the lines that I am trying to straighten and help to
                illustrate the degree to which the straightening worked.



                In particular - it appears that the line defined by CP pairs 0, 1 and 2 is
                substantially improved. However, the line defined by CP pairs 3, 4, 5, 6, 7
                and 8, while improved, is far from perfect. The barrel distortion seems to
                be much improved near the middle of the line, but not near the ends. The
                ends still curve significantly away from the reference line.



                I tried optimizing for 'b' only and I tried optimizing for 'a', 'b' and 'c'.
                I tried defining fewer CP pairs along each line and more CP pairs along each
                line. I was unable to get results better than those illustrated in the
                reference image. That said, I also tried optimizing for vertical shift (VS).
                It seems to me that the first time I tried optimizing for VS as well as for
                'b', I did get a much better result. Thinking that this might have been the
                key, I went back and defined additional CP pairs, hoping to get perfect
                results. I again optimized for 'b' and for 'VS'. This time I got 'too good
                to be true' from the optimizer and the resulting image was just a black
                canvas. I tried again and again to reproduce the promising results that I
                thought I had obtained by optimizing for 'VS' as well as for 'b', but every
                time, got 'too good to be true'.



                So - here are my questions:

                1. Am I asking for too much? Is this as good as it gets?
                2. Is it possible to optimize for VS (without having recorded how many
                mm of shift were actually used)? Could I in fact have got much better
                results on one of the optimization runs when I selected 'VS' but for some
                reason am simply unable to reproduce it?
                3. How do I optimize for VS?
                4. How do I get better results?



                Many thanks for any assistance/insight!



                BTW - on the reference image of the CP dialog box, several CPs overlap. So,
                for example, on the right pane, CPs 0 and 1 overlap and therefore, CP 0 on
                the right pane is not visible.



                Yoram

                -----Original Message-----
                From: PanoToolsNG@yahoogroups.com [mailto:PanoToolsNG@yahoogroups.com] On
                Behalf Of Erik Krause
                Sent: Friday, January 04, 2008 4:18 AM
                To: PanoToolsNG@yahoogroups.com
                Subject: RE: [PanoToolsNG] FW: calibrating a lens/teleconverter combo using
                PTGui





                Serge Maandag-3 wrote:
                >
                >> Wouldn't I then want to
                >> proceed by designating lots of subsegments along the line that I
                >> eventually
                >> want rendered straight?
                >
                > No, The three coefficients a, b and that the panotools optimizer uses can
                > only
                > correct barrel and pincushion distortion. The two line segments are enough
                > info for
                > the optimizer to do that. Subsegments would only be useful to fix a wobbly
                > line. But
                > that wouldn't be possible anyway with only a,b and c.
                >
                > The less segments you define, the less confusing it is to the optimizer.
                > If I have
                > to correct a barrel shaped box, I define a line (consisting of 2 line
                > segments) on
                > each side of the box.
                >

                Sorry, this is wrong. The use of three coefficients (a, b and c) was
                especially necessary to allow the correction of wavy (or wobbly) distortion
                as well. Wavy distortion is a mixture of barrel and pincushion distortion in
                different distances to the image center. For details see
                http://wiki. <http://wiki.panotools.org/Wavy_distortion>
                panotools.org/Wavy_distortion

                To correct for this kind of distortion you need more than two pairs of
                control points on each line and you need lines in different distances from
                the center. The optimizer will not be confused at all, since there isn't a
                possibility of false local minima.

                best regards

                -----
                Erik Krause
                http://www.erik- <http://www.erik-krause.de> krause.de
                --
                View this message in context: http://www.nabble.
                <http://www.nabble.com/RE%3A-FW%3A-calibrating-a-lens-teleconverter-combo-us
                ing--PTGui-tp14569835p14615206.html>
                com/RE%3A-FW%3A-calibrating-a-lens-teleconverter-combo-using--PTGui-tp145698
                35p14615206.html
                Sent from the PanoToolsNG mailing list archive at Nabble.com.





                [Non-text portions of this message have been removed]
              • Erik Krause
                ... Sorry for taking me so long. I missed your post because of the entirely new subject in an older thread... [...] ... No. ... Check vertical shift on the
                Message 7 of 13 , Jan 27, 2008
                • 0 Attachment
                  On Thursday, January 24, 2008 at 17:20, Yoram Bernet wrote:

                  > However, i'm still not getting the results that I hoped for. I shot a
                  > close up of a model using the Canon 24 mm tilt/shift lens. I used
                  > vertical shift. I'm trying to correct for barrel distortion.

                  Sorry for taking me so long. I missed your post because of the
                  entirely new subject in an older thread...

                  [...]
                  > 1. Am I asking for too much? Is this as good as it gets?

                  No.

                  > 2. Is it possible to optimize for VS (without having recorded how many
                  > mm of shift were actually used)? Could I in fact have got much better
                  > results on one of the optimization runs when I selected 'VS' but for some
                  > reason am simply unable to reproduce it?

                  Check "vertical shift" on the optimizer tab page and optimize again.
                  Be prepared that your result slips out of center. Hence you might
                  need to use a larger output field of view. You can adjust this in the
                  pano editor (ctrl+e). It could well be, that the PTGui optimizer
                  won't give you good results. If this is the case, use panotools
                  optimizer.

                  Using a shift lens is a classical case of center of distortion not
                  beeing in the image center. However, panotools should handle this
                  case properly, but if you use shift and lens correction optimization
                  together, the optimizer might find a false minimum. In this case you
                  might want to tune your control points and/or use more lines.

                  Please not, that if you deselect a parameter for optimization, this
                  won't reset the value. You must set these parameter manually to zero
                  on lens settings tab.

                  best regards

                  Erik Krause
                  http://www.erik-krause.de
                • Rick Drew
                  I received a few emails asking about my home-made camera warmer / jacket for shooting outdoors in winter. My current one was constructed from a fleece hat I
                  Message 8 of 13 , Jan 27, 2008
                  • 0 Attachment
                    I received a few emails asking about my home-made camera warmer / jacket for
                    shooting outdoors in winter. My current one was constructed from a fleece
                    hat I purchased at WalMart for $2.00. Cutting and sewing took about two
                    hours. I added photos to flickr.com - keyword rickscamerahat .



                    http://flickr.com/search/?q=rickscamerahat
                    <http://flickr.com/search/?q=rickscamerahat&w=22793959%40N06>
                    &w=22793959%40N06



                    Hope it helps.



                    Rick Drew



                    PS - I used it yesterday. In the morning the wind-chill was 3 degrees
                    Fahrenheit. One battery lasted for 911 photos!



                    [Non-text portions of this message have been removed]
                  Your message has been successfully submitted and would be delivered to recipients shortly.