Erratic Stitching with Ricoh Theta Z1 and PTGui 12.1...

461 views
Skip to first unread message

Brad Berson

unread,
Apr 3, 2021, 9:39:12 AM4/3/21
to PTGui Support

I'm working on setting up a work flow with the Z1 and the Dual Fisheye Plugin generating 9-shot stacked HDR-DNG , trying very hard not to have Photoshop in the middle of it and use GIMP as a last resort if at all, for touching up a stitched equirectangular TIFF from PTGui.

Following the advice to first set up a template, I took a test photo in our apartment with the camera horizontal.  Found enough control points to get a fairly decent stitch, added some more in a few blind spots, so far so good.  Saved the template, made a JPG test, looked OK.

That was with the Rnnnnnnn.DNG file from the DFE plugin.

Load up the Rnnnnnnn_HDR.DNG, apply the template, the stitch goes off the rails.

Load up both DNGs, seemingly goes OK.  Subtract the non-HDR, stitch again, stitch is screwed up again.  It looks like the seam got "pinched" somehow, even though the JPG output from PTGui (destination will be Facebook) is the same overall dimensions.

Something funny going on here.

Making things more complicated for me or for the task is that I can't generate control points with just the _HDR file loaded.  PTGui seems only to see the EV-4 data, or at least that's how it looks.  Only the most well-lit bits can be seen in the two thumbnails when I load the file.

Any guidance appreciated.  If there's someplace I can FTP the files, lemme know...

Thanks,
Brad

Erik Krause

unread,
Apr 3, 2021, 9:43:07 AM4/3/21
to pt...@googlegroups.com
Am 03.04.2021 um 15:39 schrieb Brad Berson:

> Any guidance appreciated. If there's someplace I can FTP the files, lemme
> know...

Use a file sharing service like wetransfer, dropbox etc. and post the
link here.

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

Brad Berson

unread,
Apr 3, 2021, 10:08:39 AM4/3/21
to PTGui Support

John Houghton

unread,
Apr 3, 2021, 11:19:34 AM4/3/21
to PTGui Support
Brad, I don't find the stitch "goes off the rails" with the HDR images.  There are minor differences owing to the optimum seam placement not being exactly the same, but  there's no pinching such as you describe.  All I did was load the image, apply the template, create the panorama. No control point generation or optimization.

John 

Erik Krause

unread,
Apr 4, 2021, 8:26:01 AM4/4/21
to pt...@googlegroups.com
Am 03.04.2021 um 16:08 schrieb Brad Berson:
> Thanks, here goes...
> https://www.dropbox.com/s/t16bdflat6gpmn5/PTGui-20210403001.zip?dl=0

I got the same result as John. Only when I pressed Align Images after
applying the template, the result got weird. Simply don't do that.

Joost, it seems the control point generator is still much worse for
those initially quite dark DNG HDR images. It found only three points in
the window. More control points where found only after I used Generate
Control Points Here.

Brad Berson

unread,
Apr 4, 2021, 10:25:55 AM4/4/21
to PTGui Support
Thanks both of you for looking at this.

I'm hating Google today.  I had a whole process I spelled out and Google's UI ate it.  But long story short, I cannot reproduce the problem now.  Maybe I did something stupid like hitting F5 in PTGui.  Don't know.  Except I'm getting slightly better stitches in one or two spots (like that aberration on the edges halfway up the right side TV speaker) with Ricoh's native DNGs than with the DFE DNGs using the same template.

Welcome to the Gaslight Express for me.  ;-)

The control points thing is a little frustrating.  Is there some window into the workings of the templates?  Or does it all stay under the sheets?  It's not like the points pop up after applying a template.  And trying to generate points with HDR DNGs is not so straightforward.

Thanks again,
Brad

Erik Krause

unread,
Apr 4, 2021, 11:38:36 AM4/4/21
to pt...@googlegroups.com
Am 04.04.2021 um 16:25 schrieb Brad Berson:
> The control points thing is a little frustrating. Is there some window
> into the workings of the templates? Or does it all stay under the sheets?
> It's not like the points pop up after applying a template.

Templates are standard project files, so you can use any project file as
a template. For more information read
https://www.ptgui.com/support.html#3_9 and following.

> And trying to generate points with HDR DNGs is not so straightforward.
Seems so. Problems with control point generation ind HDR and raw files
where reported long time ago and my impression was that they are fixed...

Meanwhile what I did: Shift-drag in control point editor and right click
into the rectangle. You should get "Generate Control Points Here". Click
that. If there are regions with features inside the overlap where no
points are found, press F5 to optimize. Repeat.

Brad Berson

unread,
Apr 5, 2021, 10:39:34 PM4/5/21
to PTGui Support
Thank you again, both of you.  I'm gaining a better understanding.

Nonetheless trying to get any useful number of control points from the DFE plug-in's _HDR.DNG seems largely futile.  I need to invest a little more time getting as perfect as possible a template based on an EV-0 exposure to use for my Z1 projects.

Using a template, then adding more control points seems counter-intuitive, no?

This is all somewhat different from when I started using PTGui years ago with my Canon 20D and EF10-22mm lens.  ;-)

Is there any other general advice any of you folks think I'll find useful for working with the Theta Z1 in PTGui?  Without having Photoshop in the work flow?  Looks like my brother in law wants to buy one so it'll probably be me answering his questions soon, LOL...

Thanks again,
Brad

John Houghton

unread,
Apr 6, 2021, 5:09:48 AM4/6/21
to PTGui Support
Brad, stitching images from one-shot 360 cameras with PTGui is always likely to be compromised by parallax issues to some degree, however carefully the template has been prepared.  Lens distortion and individual lens shooting angles don't change from panorama to panorama, so in theory control points and optimization should not be required to achieve good alignment.  Certainly this is true of outdoor subjects where things are generally far away and therefore relatively immune to parallax effects.  It's a different story indoors with close subjects: control points and optimization cannot correct parallax, though they might help to hide consequent misalignments along with smart seam routing and masking.  Another sensible thing to do is to carefully angle the camera so as to position the eventual seam lines in areas where misalignments are less likely to be visible - e.g. on blank walls.

If you prepare your template as described in the FAQ ( https://www.ptgui.com/support.html#6_43 ), that is probably as good as you are likely to get.  If you want to try generating your own set of lens parameters, then take two shots for an outdoor panorama, rotating the camera round through 90 degrees between them.  Load the two camera images into PTGui, which gives you four image circles to align with 50% overlap and therefore plenty of features for control points and optimization.   Remove one image pair from the project before saving as a template.

John

D2R Survey

unread,
Oct 23, 2023, 9:44:54 AM10/23/23
to PTGui Support
Hi

I am trying to use the Z1 to colourise laser scan data from a ZF 5010.  This basically involves presenting a pano of matching resolution to the scan from the same position.  In practice, take scan, move to next tripod, take pano shots and so on.  I can do this with a DSLR and nodal ninja rig, but this takes 6 shots.  The Z1 us quicker even if I take 2 per position at 90 deg.  Having stitched a project using the Lightroom Theta plugin, I can see parallax issues where scan data is coloured incorrectly because the image doesn't match the exact scanned scene.  So I am here in PTGui wondering if I can get a better stitch, taking 2 shots per position.  Effectively a calibration per scene.

Faro scanners have adopted this camera, so one would assume it is possible.  Their routine is to run a calibration before sets of scans.

Is it better to create one template with an outdoor scene as described, or take 2 shots per position and create the pano from both double shots?  I am struggling with the exact workflow in PTgui having presented it two images.  1 image per set was automatically masked, which I have removed here.  I am trying to find the best, most accurate workflow to see if this camera is viable to use.

Screenshot 2023-10-23 090906.png
 
The created pano looks great, except directly underneath the camera.  I don't usually have this area on a tilted DSLR rig and nor does the scanner measure this area, it is in the blind spot.  The floor would be captured from the adjacent scan.

Screenshot 2023-10-23 091908.png

Thanks

PTGui Support

unread,
Oct 23, 2023, 10:22:47 AM10/23/23
to pt...@googlegroups.com
Hi,

If you use 2 back-to-back images (like those from the Z1 or any dual
fisheye camera), PTGui will try to match the control points as good as
possible, so the resulting panorama will be as seamless as possible in
the area where the two images overlap. But the overlap is just a tiny
cylindrical area, and anything outside the overlap area may not match
your laser scan image at all. To overcome this, you would need to take 2
(x2 fisheye) images with the camera rotated 90 degrees (so you have an
image at 0, 90, 180 and 270 degrees). The overlap area will now span the
entire panorama, but then you could probably just as well use a SLR on a
tripod.

For the best match I guess it's best if you load the individual images
together with the laser scan image in a PTGui project, and add some
control points between the laser scan and the fisheye images. The
process is outlined in the second part of 6.27:
https://ptgui.com/support.html#6_27

I'm not familiar with laser scanners, but you should also consider
whether the laser scan data suffers from parallax. If either image
suffers from parallax you'll never get a perfect match between the two.

Joost

On 23-10-2023 14:49, D2R Survey wrote:
> Hi
>
> I am trying to use the Z1 to colourise laser scan data from a ZF 5010.
> This basically involves presenting a pano of matching resolution to the
> scan from the same position.  In practice, take scan, move to next
> tripod, take pano shots and so on.  I can do this with a DSLR and nodal
> ninja rig, but this takes 6 shots.  The Z1 us quicker even if I take 2
> per position at 90 deg.  Having stitched a project using the Lightroom
> Theta plugin, I can see parallax issues where scan data is coloured
> incorrectly because the image doesn't match the exact scanned scene.  So
> I am here in PTGui wondering if I can get a better stitch, taking 2
> shots per position.  Effectively a calibration per scene.
>
> Faro scanners have adopted this camera, so one would assume it is
> possible.  Their routine is to run a calibration before sets of scans.
>
> Is it better to create one template with an outdoor scene as described,
> or take 2 shots per position and create the pano from both double shots?
> I am struggling with the exact workflow in PTgui having presented it two
> images.  1 image per set was automatically masked, which I have removed
> here.  I am trying to find the best, most accurate workflow to see if
> this camera is viable to use.
>
> Screenshot 2023-10-23 090906.png
>
> The created pano looks great, except directly underneath the camera.  I
> don't usually have this area on a tilted DSLR rig and nor does the
> scanner measure this area, it is in the blind spot.  The floor would be
> captured from the adjacent scan.
>
> --
> 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 view this discussion on the web visit
> https://groups.google.com/d/msgid/ptgui/d7f3a641-bb00-4590-86cf-46542c8c502en%40googlegroups.com <https://groups.google.com/d/msgid/ptgui/d7f3a641-bb00-4590-86cf-46542c8c502en%40googlegroups.com?utm_medium=email&utm_source=footer>.

D2R Survey

unread,
Oct 23, 2023, 11:19:05 AM10/23/23
to PTGui Support
Hi, thanks for your reply.

Scan data we are going to take as accurate XYZ coordinates, each point has an intensity value which we is converted to grey scale.  The colour pixels information replaces the grey scale.  Laser Control software does that matching, but I need to offer it the right pano.

So here we can see the red cone has grey concrete, as do the pallets.  This is not a good stitch.   10000 x 5000 res scan and pano

Screenshot 2023-10-23 160205.png

This scene is near perfect in that the light coloured wood edge is not mapped to the floor or vice versa.  5000 x 2500 res scan and pano

Screenshot 2023-10-23 160746.png

I was wondering if I am stretching the pano as I need to up width and height to 10000 x 5000 for high-res scans?

Screenshot 2023-10-23 161030.png

Even if I had to take 3 per position, this is still far easier than using the DSLR rig.

Screenshot 2023-10-23 161420.png

I am trying to find the absolute best practice workflow through PTgui.  1 shot stitching I have found hit-and-miss so hence trying the two shot method.

Regards'

PTGui Support

unread,
Oct 23, 2023, 11:58:31 AM10/23/23
to pt...@googlegroups.com
I'm not sure I understand your workflow, but it looks like there's
something wrong with it. The orange cone matches at ground level, but
the top doesn't match. This looks as if they were taken from different
viewpoints?

Kind regards,

Joost Nieuwenhuijse
www.ptgui.com

On 23-10-2023 17:19, D2R Survey wrote:
> Hi, thanks for your reply.
>
> Scan data we are going to take as accurate XYZ coordinates, each point
> has an intensity value which we is converted to grey scale.  The colour
> pixels information replaces the grey scale.  Laser Control software does
> that matching, but I need to offer it the right pano.
>
> So here we can see the red cone has grey concrete, as do the pallets.
> This is not a good stitch.   10000 x 5000 res scan and pano
>
> Screenshot 2023-10-23 160205.png
>
> This scene is near perfect in that the light coloured wood edge is not
> mapped to the floor or vice versa.  5000 x 2500 res scan and pano
>
> Screenshot 2023-10-23 160746.png
>
> I was wondering if I am stretching the pano as I need to up width and
> height to 10000 x 5000 for high-res scans?
>
> Screenshot 2023-10-23 161030.png
>
> Even if I had to take 3 per position, this is still far easier than
> using the DSLR rig.
>
> https://groups.google.com/d/msgid/ptgui/d7f3a641-bb00-4590-86cf-46542c8c502en%40googlegroups.com <https://groups.google.com/d/msgid/ptgui/d7f3a641-bb00-4590-86cf-46542c8c502en%40googlegroups.com> <https://groups.google.com/d/msgid/ptgui/d7f3a641-bb00-4590-86cf-46542c8c502en%40googlegroups.com?utm_medium=email&utm_source=footer <https://groups.google.com/d/msgid/ptgui/d7f3a641-bb00-4590-86cf-46542c8c502en%40googlegroups.com?utm_medium=email&utm_source=footer>>.
>
> --
> 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 view this discussion on the web visit
> https://groups.google.com/d/msgid/ptgui/dbb3eb9d-a678-4613-a78d-78e1530f202cn%40googlegroups.com <https://groups.google.com/d/msgid/ptgui/dbb3eb9d-a678-4613-a78d-78e1530f202cn%40googlegroups.com?utm_medium=email&utm_source=footer>.

D2R Survey

unread,
Oct 23, 2023, 12:17:06 PM10/23/23
to PTGui Support
This is a scanned scene with colour mapped points from a pano.  This process is not unlike creating a real-estate virtual tour, however scan data is exact, so we can extract dimensions.  It is used for architectural design purposes.  I measure and draw existing buildings.  Colour isn't strictly necessary, however it is useful.  So here we have presented the pano for mapping, but clearly the pano isn't true as we have red bollard on the floor.  I am trying to get through to the best Ptgui workflow to produce the least distorted pano. 

Screenshot 2023-10-23 170334.png

Regards'
Reply all
Reply to author
Forward
0 new messages