PTGui is so good it patches the zenith and nadir automatically...
Ken Warner
why.
OK, with a fisheye -- 4 around at 7 degrees down -- camera zoomed
slightly to fill the frame so that the sides in portrait mode
are just clipped a small amount leaving the top and bottom
full circle.
I cropped to the inside of the usual fisheye rolloff zone where
the image begins to dim.
Usually, this setup will leave a small hole at the zenith and
I will see the top of my NN SPH2 on my monopod. Usually I would
see some parts of my hand wrapped around the monopod underneath the
top of the SPH2.
Well, in this unretouched pano I made today, the zenith is patched
and the nadir somehow got a down shot patch -- must be magic.
When I align the images, PTGui reports back that the horizontal
FOV is 189 degrees which is impossible. It should be much less than
180 degrees. Probably around 165-170 degrees.
I've attached the pts file and here's a link to the image and one
in my viewer.
http://pancyl.com/images/Motel.jpg
Now I shouldn't really complain because automagically patched zenith
and nadir shots are what we'd all like right?
Ken
John Houghton
> Now I shouldn't really complain because automagically patched
> zenith and nadir shots are what we'd all like right?
we don't have that here. The lens parameters have optimized to
unlikely values, which accounts for the expanded angle of view that
closes up the zenith and nadir holes. Try to improve the vertical
spread of control points by assigning some manually. Better still,
perform a calibration exercise to evaluate a proper set of lens
parameters. There is also evidence of parallax problems, which will
also not help the optimizer to evaluate good lens parameters.
John
Ken Warner
I know it's not accurately stitched -- that's why the message :-)
Parallax is always a problem when using a monopod. I figure if
I can rotate the nodal point within a 1-2 inch circle, I'm doing
good. And in close quarters like that walkway, glitches in stitching
are guaranteed. I can usually work those out in post. I left
the image untouched just to clearly indicate the problem.
Lens calabration is important but given that I zoom the camera
lens slightly to fill the frame with the FE circle and that
zoom level is not always the same, calabration would have to
be done for each possible zoom level.
What I'm looking for is a good enough workaround so that
the calabration that PTGui performs is good enough for a reasonable
stitch. PTGui usually is good enough. This particular image set
has unique problems that I'm not good enough with PTGui to work
through.
I think there is a method to coax PTGui into doing a reasonable stitch
on this images set. I'm just looking for that method.
Ken
Bjørn K Nilssen
> What I'm looking for is a good enough workaround so that
> the calabration that PTGui performs is good enough for a reasonable
> stitch. PTGui usually is good enough. This particular image set
> has unique problems that I'm not good enough with PTGui to work
> through.
>
> I think there is a method to coax PTGui into doing a reasonable stitch
> on this images set. I'm just looking for that method.
The method I'm using if PTgui insists on some unreasonable fov values is to just set the
fov myself to something like 183-185, which is what I usually get on good stitches. It's
a nice KISS solution that works fine ;)
BTW, I also zoom in to crop off along the long sides, but as long as I use the circular
crop/lens PTgui apparently calculates the fov of the circle, and not the actual cropped
frame sides.
--
Bjørn K Nilssen - http://bknilssen.no - panoramas and 3D
Ken Warner
I did two things that makes a better stitch.
1) I was optimizing the Field of View. I turned that off
and Restored Default lens correction parameters to 0,0,0 and
then set the Horizontal field of view to 177 which is about
right -- then I optimized and got the hole in the zenith and
some of the top of the monopod -- which is an improvement.
2) I moved the crop circle outside the actual image circle and deleted
all control points and aligned and optimized again. That
also helped to produce a more realistic optimization.
Then I deleted all control points greater than 10.0 and I think
that gives me a good enough stitch that I can post process it to
take out the glitches.
It's the optimization of the Field of View that messed up the
stitch of this set of images.
Ken
Ken Warner
http://pancyl.com/images/Motel.jpg
http://pancyl.com/Motel.htm
John Houghton
> Lens calabration is important but given that I zoom the camera
> lens slightly to fill the frame with the FE circle and that
> zoom level is not always the same, calabration would have to
> be done for each possible zoom level.
circle on the actual image circle, the calibrated image parameters
will remain reasonably valid - certainly preferable to grotesque ones
generated by an automatic process. If you avoid control points on
near features and include only fov and possibly h&v shift parameters
in the optimization, the overall alignment of the panorama should be
satisfactory, though stitching errors on near features are to be
expected, and will need attention in post-processing. I prefer this
to using the optimizer to correct parallax errors, which tends to lead
to unwelcome general distortions.
John
Ken Warner
Your method sounds like it is best applied with a template. I
agree -- in theory -- that this is the best way. But with
my equipment, the setup is always slightly different from
the last time.
The zoom level changes. The position of the camera on the NN2/3
changes. Just getting the crop circle in the same place is
next to impossible.
Suggestion: for the Crop Tab workspace -- allow the background to
be changed to a different color. Black makes it too hard to see
the edge of the FE circle because of the rolloff to black on the edges
of the FE circle. Also, support the Arrow Keys for
positioning the crop circle one pixel -- like CP's. And
maybe even support a grid on the crop window. And draw
a small cross hair in the middle of the crop circle. And
use the +/- keys to size the crop circle on pixel at a time.
The lens on my compact camera actually wiggles around. The
image circle is never in the same place from image set to image
set.
You suggest optimizing the FOV. But that seemed to be causing
most of the problems I was having with this set of images. Was
that happening because I was using automatically generated distortion
params for the lens? So if I generate a good set of distortion
params and save them in the lens DB and then when they are applied,
optimize the FOV but not the distortion params? That sounds like
a good thing to do. I'll work that into my method.
There is always a temptation to put CP's on near ground objects
because that's where you see the most parallax error. I try
to avoid that temptation and I've found that if PTGui puts
CP's there, removing them helps.
It would be good to get some consistency in my method but each
image set seems different from the last. Hopefully I will become
more consistent.
Ken
Andrew Stephens
I'm responding off-list because I want to talk about Autopano Pro and
that's not fair on a PTGui email list!
I want to suggest you try stitching these FE images with the new
version 2.01 of Autopano Pro or Auopano Giga - they've done a lot of
work on FE circle detection and FE stitching and it's working very
well indeed for me.
http://www.autopano.net/en/buy-autopano/download.html
I'd be interested in hearing how it work for you.
I'd also love to have a go at stitching some of your image sets if
you are willing to share them.
Regards,
Andrew (aka mediavets)
.............
>Checked by AVG - www.avg.com
>Version: 8.5.325 / Virus Database: 270.12.31/2116 - Release Date:
>05/15/09 06:16:00
Mike Johnston
It went to the group.
Mike J
Andrew Stephens
Yet another 'senior moment'.
Oh dear....
Andrew
.........
Ken Warner
It would be interesting to see how APP does on difficult image
sets like the kind I seem to be able to make :-)
I don't want to set up APP right now but here's a half size set
of the image I posted. A full size set is over 80 meg. I have
a slow connection. They should be crummy enough for a good test.
I'd like to see the results. It's 4 around, 1 up, 1 down.
http://pancyl.com/images/Motel.zip
Ken