Contorted horizontal seam lines

90 views
Skip to first unread message

Tom Sharpless

unread,
Mar 18, 2018, 4:16:10 PM3/18/18
to PTGui Support
PTGui often generates weird horizontal seam lines on two-row panoramas. 
Sometimes they even loop back over themselves, as in the example below (two rows of 6-around, shot from monopod, auto-aligned, PTGui 10.0.17).
This is easily corrected with a bit of masking, but I would prefer not to have to do that.


Erik Krause

unread,
Mar 18, 2018, 4:31:11 PM3/18/18
to pt...@googlegroups.com
Am 18.03.2018 um 21:16 schrieb Tom Sharpless:
> PTGui often generates weird horizontal seam lines on two-row panoramas.
> Sometimes they even loop back over themselves, as in the example below (two
> rows of 6-around, shot from monopod, auto-aligned, PTGui 10.0.17).

Funny! Is this a full frame fisheye? I always use full frame fisheyes as
circular ones, with a crop circle diameter equal to the longer side.
This way I get rid of the corners which are horn-like in equirectangular
projection and extend too far into neighboring images and might well
cause this strange seam lines.

--
Erik Krause
http://www.erik-krause.de

Tom Sharpless

unread,
Mar 18, 2018, 4:41:55 PM3/18/18
to PTGui Support
Yes, Erik, full frame (Samyang 8mm on APS-C camera).
I like your idea to ask PTGui to crop this type of image, and will be using it.  
That cuts off the rather useless corners, and also makes "field of view" used for lens calibration explicitly visible.

Tom Sharpless

unread,
Mar 18, 2018, 4:58:42 PM3/18/18
to PTGui Support
But cropping still does not give straight seams. 
Same pano with crop circle just touching top & bottom of sensor rectangle.
The loops are gone but there is still a vertical "singularity" at the center of each view.

PTGui Support

unread,
Mar 19, 2018, 3:22:20 AM3/19/18
to pt...@googlegroups.com
The seams are calculated by finding the point in the overlapping images
which is closest to the center of an image. Apparently this results in
these seam lines. I've never seen those loops though.

Kind regards,

Joost Nieuwenhuijse
www.ptgui.com

On 18/03/2018 21:58, Tom Sharpless wrote:
> But cropping still does not give straight seams.
> Same pano with crop circle just touching top & bottom of sensor rectangle.
> The loops are gone but there is still a vertical "singularity" at the
> center of each view.
>
> <https://lh3.googleusercontent.com/-kwUF5tppuzg/Wq7SV5XSpyI/AAAAAAAACOs/bu5vAsj512Qbpy0uRYvLoRnxK4czz-QNgCLcBGAs/s1600/ContortedSeams-crop.jpg>
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To post to this group, send email to pt...@googlegroups.com
> <mailto:pt...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/ptgui.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/ptgui/00d00cb9-471b-42e6-b5f2-3b39455c62b9%40googlegroups.com
> <https://groups.google.com/d/msgid/ptgui/00d00cb9-471b-42e6-b5f2-3b39455c62b9%40googlegroups.com?utm_medium=email&utm_source=footer>.
> For more options, visit https://groups.google.com/d/optout.

Erik Krause

unread,
Mar 19, 2018, 4:35:14 AM3/19/18
to pt...@googlegroups.com
Am 19.03.2018 um 08:22 schrieb PTGui Support:
> The seams are calculated by finding the point in the overlapping images
> which is closest to the center of an image. Apparently this results in
> these seam lines. I've never seen those loops though.

I can reproduce the loops. It comes if roll is different for two images
vertically above each other. F.e. both images: Yaw 0, first: Pitch 15,
Roll 7, second: Pitch -15, Roll 0.

It gets pretty weird if you have Roll 90 and 0.

John Houghton

unread,
Mar 19, 2018, 5:12:21 AM3/19/18
to PTGui Support
Simply setting the same 1 degree roll for all of the images has a profound effect on the horizontal seam line, and the screenshot indicates around 3 degrees of roll.

John

Tom Sharpless

unread,
Mar 21, 2018, 11:24:01 AM3/21/18
to PTGui Support
Of course in a real stitch, adjusting roll values.is not an option.
I can't say that this anomaly has ever caused me trouble in stitching a regular panorama, and as Erik points out it can be reduced by applying a circular crop.

But there is a related anomaly that I consider a real problem.  This happens when I add masks to define the seam line I want, usually above or below the midline where PTGui puts the seam by default.  After adding simple red masks over the bottom of the upper row and the top of the lower row, the seam lines look as shown in the attached image.  There are triangular "loops" where the upper image comes down over the lower one (or vice versa) in a most unexpected fashion.  If I add a green dot  next to each red mask, the loops disappear and I get the intended straight horizontal seam.   As this is an uncorrected stereo stitch, you can see considerable misalignment between the top and bottom images.

For stereo and hand-held stitching I use a warp-to-fit method implemented by my helper programs PT3D (stereo) and PTMP (mono)  This does a good job of eliminating stitch errors visible at the seams, without visibly distorting the panorama; but only if all control points are placed near seams.  So it is critical that the helper can predict where PTGui will put the seams.  I have not been able to reproduce PTGui's way of choosing seam lines,  nor, in view of various anomalies, would I really want to.   I have been increasingly relying on having the helpers create masks that force seams to known positions.  Just now I am equipping PT3D with the ability to mask two-row panoramas in the way described above.  I expect this to be a big time saver for me as I prefer to shoot two rows with a full-frame fish-eye.

PTGui Support

unread,
Mar 21, 2018, 12:18:56 PM3/21/18
to pt...@googlegroups.com
It's pretty straightforward: for each pixel in the panorama PTGui
examines the source images which contribute to that pixel in the
panorama. It uses the pixel from the source image where it has the
greatest distance from a red mask or from the edge. In other words the
seam is moved as far away from all image edges and from all red masks.

Kind regards,

Joost Nieuwenhuijse
www.ptgui.com

> <https://lh3.googleusercontent.com/-FvI_IxWsXyM/WrJ4-9yQ2kI/AAAAAAAACPE/ngWuUuYa5GoI33-HAKeW7cvk6JJxFYNZQCLcBGAs/s1600/seam-triangles.jpg>
>
>
>
> On Monday, March 19, 2018 at 5:12:21 AM UTC-4, John Houghton wrote:
>
> On Monday, March 19, 2018 at 8:35:14 AM UTC, Erik Krause wrote:
>
> Am 19.03.2018 um 08:22 schrieb PTGui Support:
> > The seams are calculated by finding the point in the
> overlapping images
> > which is closest to the center of an image. Apparently this
> results in
> > these seam lines. I've never seen those loops though.
>
> I can reproduce the loops. It comes if roll is different for two
> images
> vertically above each other. F.e. both images: Yaw 0, first:
> Pitch 15,
> Roll 7, second: Pitch -15, Roll 0.
>
> It gets pretty weird if you have Roll 90 and 0.
>
>
> Simply setting the same 1 degree roll for all of the images has a
> profound effect on the horizontal seam line, and the screenshot
> indicates around 3 degrees of roll.
>
> John
>
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To post to this group, send email to pt...@googlegroups.com
> <mailto:pt...@googlegroups.com>.
> Visit this group at https://groups.google.com/group/ptgui.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/ptgui/923cad95-db14-42ac-8afe-6726a11f2b06%40googlegroups.com
> <https://groups.google.com/d/msgid/ptgui/923cad95-db14-42ac-8afe-6726a11f2b06%40googlegroups.com?utm_medium=email&utm_source=footer>.

Thomas Sharpless

unread,
Mar 21, 2018, 4:12:33 PM3/21/18
to pt...@googlegroups.com
What about green masks?


To unsubscribe from this group and stop receiving emails from it, send an email to ptgui+unsubscribe@googlegroups.com <mailto:ptgui+unsubscribe@googlegroups.com>.

--
You received this message because you are subscribed to a topic in the Google Groups "PTGui Support" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ptgui/8m0NrIqVO5o/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ptgui+unsubscribe@googlegroups.com.
To post to this group, send email to pt...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages