Front/Back Parallax With Sigma 15mm f/2.8
Kevin Morrissey
I'm using the following combination of equipment/software: Canon EOS
5D body with Sigma 15mm f/2.8 [full-frame] fisheye; Bogen 303+
Panoramic Head with leveling base. With the current settings I'd
found for the left-right and front-back standards on the 303+ head, I
had heretofore had good luck with shooting church interiors.
I should add that this rig is on top of an eleven foot stand in the
center of the interior, to reduce "keystoning" in architectural
features such as windows, etc.
For the sake of expedience I used a small tripod to shoot a 360° X
180° of a church interior recently, and encountered all sorts of
parallax problems that I thought I'd eliminated.
My question is this: would you concur with me that when the 15mm f/
2.8 is eleven feet up in the air in the center of the building, both
"keystoning" and parallax are reduced because everything I'm focusing
on is closer to infinity?
(I'm not sure whether a small attachment will fit into this post; here
it is if it will):
/Users/kevinmorrissey/Desktop/sh.gif
Anyway, I'd really appreciate any input you could give; PT Gui is of
course in a class of its own and has provided me with a lot of
positive challenge and inspiration.
Thanks for your time and attention!
Kevin Morrissey
annean...@comcast.net
Keith Martin
> 2.8 is eleven feet up in the air in the center of the building, both
> "keystoning" and parallax are reduced because everything I'm focusing
> on is closer to infinity?
As for keystoning, that is a matter of perspective and angle/point of view. Shooting from lower down will increase this, whatever kit you use. (Other than tilt/shift lens gear of course, but that's not recommended for pano work.)
k
John Houghton
I don't think the 303+ is a spherical head, so if you are making 360
cylindrical panoramas, keystoning should not be an issue. Perhaps you
could explain what sort of panoramas you are making and if these are
for printing or viewing in a Quicktime or Flash viewer.
The 303+ seems an awkward head to set up accurately. How are you
checking the lateral position to centralize the entrance pupil?
John
Anne and Kevin Morrissey
--
You received this message because you are subscribed to the Google Groups "PTGui" group.
To post to this group, send email to pt...@googlegroups.com
To unsubscribe from this group, send email to ptgui+un...@googlegroups.com
Please do not add attachments to your posts; instead you may upload files at
http://groups.google.com/group/ptgui/files
For more options, visit this group at http://groups.google.com/group/ptgui
To unsubscribe from this group, send email to ptgui+unsubscribegooglegroups.com or reply to this email with the words "REMOVE ME" as the subject.
Anne and Kevin Morrissey
For the prompt reply. I did upload a file but only got a little
square section of it; maybe you know a fix for that.
The 303+ is a 360° X 180°, not a spherical head; I'm going for print,
not web. I was able to pretty much eliminate keystoning with an
eleven foot camera stand; I believe not only the keystoning but also
rather extreme parallax (remember, I'm using a Sigma 15mm f/2.8) is
an issue.
With regard to checking the lateral position to centralize the
entrance pupil, I hung a thread in front of the rig and "eyeballed"
the left/right standard to bisect the lens.
With regard to the front/back standard, by the way, I used test
targets at each 5mm of front/back adjustment to arrive at a "nulled"
position.
It simply strikes me that from up on top of my tall stand (in the
center of the church), I'm focusing at infinity -- or closer to it
anyway -- thereby avoiding parallax issues.
Which projection should I be using with the Sigma 15mm f/2.8, by the
way (it's their full-frame fisheye; the 8mm f/3.5 projects a circular
image but I opted for the 15mm for two reasons: 1. It's f/2.8; 2.
It uses all of the 24X36mm sensor on the Canon 5D I paid so much
money for!
Again, thanks for your prompt response, and, of course, I love using
PT Gui!
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui" group.
> To post to this group, send email to pt...@googlegroups.com
> To unsubscribe from this group, send email to ptgui
> Please do not add attachments to your posts; instead you may upload
> files at
> http://groups.google.com/group/ptgui/files
> For more options, visit this group at http://groups.google.com/
> group/ptgui
>
> To unsubscribe from this group, send email to ptgui
Anne and Kevin Morrissey
--
You received this message because you are subscribed to the Google Groups "PTGui" group.
To post to this group, send email to pt...@googlegroups.com
To unsubscribe from this group, send email to ptgui+un...@googlegroups.com
Please do not add attachments to your posts; instead you may upload files at
http://groups.google.com/group/ptgui/files
For more options, visit this group at http://groups.google.com/group/ptgui
To unsubscribe from this group, send email to ptgui+unsubscribegooglegroups.com or reply to this email with the words "REMOVE ME" as the subject.
You received this message because you are subscribed to the Google Groups "PTGui" group.
To post to this group, send email to pt...@googlegroups.com
Please do not add attachments to your posts; instead you may upload files at
http://groups.google.com/group/ptgui/files
For more options, visit this group at http://groups.google.com/group/ptgui
Anne and Kevin Morrissey
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui" group.
> To post to this group, send email to pt...@googlegroups.com
> To unsubscribe from this group, send email to ptgui
> Please do not add attachments to your posts; instead you may upload
> files at
> http://groups.google.com/group/ptgui/files
> For more options, visit this group at http://groups.google.com/
> group/ptgui
>
> To unsubscribe from this group, send email to ptgui
> +unsubscribegooglegroups.com or reply to this email with the words
> "REMOVE ME" as the subject.
--
You received this message because you are subscribed to the Google
Groups "PTGui" group.
To post to this group, send email to pt...@googlegroups.com
To unsubscribe from this group, send email to ptgui
Please do not add attachments to your posts; instead you may upload
files at
http://groups.google.com/group/ptgui/files
For more options, visit this group at http://groups.google.com/group/
ptgui
To unsubscribe from this group, send email to ptgui
John Houghton
<anneandke...@comcast.net> wrote:
> Thanks John,
>
> For the prompt reply. I did upload a file but only got a little
> square section of it; maybe you know a fix for that.
You should upload the file in jpeg format - not indexed color, tiff.
> The 303+ is a 360° X 180°, not a spherical head;
Not sure what you mean. 360x180 usually means spherical.
> I was able to pretty much eliminate keystoning with an
> eleven foot camera stand; I believe not only the keystoning but also
> rather extreme parallax (remember, I'm using a Sigma 15mm f/2.8) is
> an issue.
There is no extreme parallax. Your head is set up adequately to avoid
extreme effects. Your problem is very poor stitching, with the lens
distortion not corrected properly.
> With regard to checking the lateral position to centralize the
> entrance pupil, I hung a thread in front of the rig and "eyeballed"
> the left/right standard to bisect the lens.
Probably good enough.
> With regard to the front/back standard, by the way, I used test
> targets at each 5mm of front/back adjustment to arrive at a "nulled"
> position.
With something very near the camera (40cm, say) set against a distant
background, you should be able to minimize parallax to 1mm of front to
back adjustment.
> It simply strikes me that from up on top of my tall stand (in the
> center of the church), I'm focusing at infinity -- or closer to it
> anyway -- thereby avoiding parallax issues.
That is of no relevance. Parallax results from the (apparent) shift
in position of the entrance pupil between shots. This is the same
whether the camera is on a tall or short tripod. Admittedly, there
may be small differences with the near pews, but these are tiny
compared to the huge misalignments in your sample image.
> Which projection should I be using with the Sigma 15mm f/2.8, by the
> way (it's their full-frame fisheye;
Use fullframe, but circular can also be used if you set the crop
circle on the corners of the images. (Some people set a smaller crop
circle to eliminate the corners, which are not normally needed).
Your main problem seems to be poor stitching. Check that you have a
good spread of control points vertically along the overlap areas.
Include the lens shift parameters (d & e) in the optimization. (In the
optimizer's Simple mode, select Heavy + Lens Shift in the Minimize
Lens Distortion field). When the images are well aligned, there will
be no keystoning, regardless of which tripod you use. If you cannot
get a good alignment, I would be happy to try a stitch if you care to
supply the set of images.
John
Hans
On Mar 29, 7:25 pm, Anne and Kevin Morrissey
<anneandke...@comcast.net> wrote:
> Thanks John,
>
> For the prompt reply. I did upload a file but only got a little
> square section of it; maybe you know a fix for that.
>
> The 303+ is a 360° X 180°, not a spherical head; I'm going for print,
> not web. I was able to pretty much eliminate keystoning with an
> eleven foot camera stand; I believe not only the keystoning but also
> rather extreme parallax (remember, I'm using a Sigma 15mm f/2.8) is
> an issue.
>
> With regard to checking the lateral position to centralize the
> entrance pupil, I hung a thread in front of the rig and "eyeballed"
> the left/right standard to bisect the lens.
>
> With regard to the front/back standard, by the way, I used test
> targets at each 5mm of front/back adjustment to arrive at a "nulled"
> position.
>
> It simply strikes me that from up on top of my tall stand (in the
> center of the church), I'm focusing at infinity -- or closer to it
> anyway -- thereby avoiding parallax issues.
Sorry but parallax is not just about front and back objects. It is
just as much about perspective.
This sis important to understand.
I do not understand your position settings for the Manfrotto head.
The upper arm should be measured from your tripod thread to the
rotation point.
For the Sigma 15mm on a 5D this should be at 80mm plus minus 1 mm.
For the lateral position you can just look through the viewfinder or
use the lifeview om Mark II and place the center focuspoint to the
rotator centre.
But of course you need a spherical head for this and I am not sure
what you have. 303+ is not anything I heard of. It is called 303 or
303 SPH
The projection does not matter, You can use fullframe or circular
projection.
There are a couple of important things,
For best depth of filed you should focus at the hyperfocal point which
at f8 is at 1m.
Remember that the focusplane is flat so you need a very large Depth of
filed eve if you are high up on a pole.
The areas at the corners will still be just 0.5m away in
focusdistance.
The best way to use the 15mm is to shoot 6 around tilted down 10
degrees and 1 zenith at 90 degree or if you have no features at the
top you can shoot at 70 degree for better controlpoint generation.
It is very important that you use advanced optimizing and optimize for
the shift parameters d+e.
Also remember to not use autorotation in camera but rotate the images
in the Raw converter instead.
Hans
>
> Which projection should I be using with the Sigma 15mm f/2.8, by the
> way (it's their full-frame fisheye; the 8mm f/3.5 projects a circular
> image but I opted for the 15mm for two reasons: 1. It's f/2.8; 2.
> It uses all of the 24X36mm sensor on the Canon 5D I paid so much
> money for!
>
> Again, thanks for your prompt response, and, of course, I love using
> PT Gui!
>
> Kevin
> > For more options, visit this group athttp://groups.google.com/
> > group/ptgui
>
> > To unsubscribe from this group, send email to ptgui
> > +unsubscribegooglegroups.com or reply to this email with the words
> > "REMOVE ME" as the subject.
>
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui" group.
> To post to this group, send email to pt...@googlegroups.com
> To unsubscribe from this group, send email to ptgui
> +unsub...@googlegroups.com
> Please do not add attachments to your posts; instead you may upload
> files athttp://groups.google.com/group/ptgui/files
> For more options, visit this group athttp://groups.google.com/group/
Anne and Kevin Morrissey
By 360° X 180°, I thought I meant simply, not spherical; intended for
the printed page, but including 360° left to right and 180°
(diagonal) top to bottom.
Is it possible to simply shoot too many frames? I shot twelve, and I
might have gotten by with ten or eight. I appreciate your offer!
But I don't want to impose to that extent just yet.
Kevin
mick crane
Anne and Kevin Morrissey wrote:
> Is it possible to simply shoot too many frames? I shot twelve, and I
> might have gotten by with ten or eight. I appreciate your offer!
> But I don't want to impose to that extent just yet.
It's probably less of an imposition than you think and I would take
him up on the offer. maybe he will see something that will save you
hours of grief.
cheers
Mick
Anne and Kevin Morrissey
For the prompt and detailed response.
The 303+ is the next head "down" from the 303SPH in Bogen's line-up;
it has front-back and left-right standards but cannot be positioned
at any angle except 90° relative to the plane of rotation: it will
create a full 360° field of view laterally, and renders as much of a
top-to-bottom FOV as the focal length of the lens in use, but it is
not designed with a plate or with click-stops to "tilt" the camera up
or down, (for true "spherical" panoramas).
To check my front/back nodal point positioning, I used close/far test
targets, positioned first at the left side, then at the right side,
of the frame. I arrived at the correct front/back setting by
comparing for the exact alignment of these targets. I understand
from John Houghton, though, that the front target should be at 40 cm,
which is a lot closer than the target I used.
Is it possible simply to shoot too many frames (i.e., with too much
-- say, 35% or 50% overlap? I think that might have had something to
do with my problem.
Thanks again for your prompt response; it's really appreciated!
Kevin Morrissey
annean...@comcast.net
> For more options, visit this group at http://groups.google.com/
John Houghton
<anneandke...@comcast.net> wrote:
> Is it possible to simply shoot too many frames? I shot twelve, and I
> might have gotten by with ten or eight.
12 is overkill. Given that the horizontal fov would be around 94
degrees, 6 shots is normally considered adequate (36% overlap). Try
stitching the alternate images.
John
Anne and Kevin Morrissey
Thanks!
K
Anne and Kevin Morrissey
where the problem arose. I'm shooting test shots in the living room
and have arrived at 8 (I've got the camera in the vertical (portrait)
position, by the way.
Thanks again, John!
P.S. Perhaps I will upload some little tiffs for you to try out.
The support is really appreciated! K
kevin
John Houghton
alignment or better quality or what?
The problems with 12 shots around may have been that because of the
very large overlap, the control point generator is likely to generate
points between non-neighboring images. I.e. between images 0 and 2
(and even 0 and 3) as well as between 0 and 1. It's best if you take
positive steps to avoid these unnecessary points. One way is to step
through pairs of images 0-1, 1-2 etc. on the Control Points tab and
use ctrl+shift+G to generate control points between the displayed
images. If you have a good spread of points and the optimizer gets
the average distance down to <1, with a maximum of <3, say, you should
find that the images align very well.
John
On Mar 30, 1:54 am, Anne and Kevin Morrissey
> > For more options, visit this group athttp://groups.google.com/
> > group/ptgui
>
> > To unsubscribe from this group, send email to ptgui
> > +unsubscribegooglegroups.com or reply to this email with the words
> > "REMOVE ME" as the subject.- Hide quoted text -
>
> - Show quoted text -
Hans
On Mar 30, 2:54 am, Anne and Kevin Morrissey
<anneandke...@comcast.net> wrote:
> Been there done that. I'm quite certain now that that's primarily
> where the problem arose. I'm shooting test shots in the living room
> and have arrived at 8 (I've got the camera in the vertical (portrait)
> position, by the way.
6 is more than enough.
Actually you could shoot 5 but most rotators does not support
clickstops for 5.
5 images gives you an overlap of 24% which is very good. 25% is an
optimal overlap which we use as a start point.
People who use the 8 mm fisheye shoot 4 images around and on a Canon
APS camera that is just 18% overlap.
Hans
>
> Thanks again, John!
>
> P.S. Perhaps I will upload some little tiffs for you to try out.
> The support is really appreciated! K
> kevin
> On Mar 29, 2010, at 4:31 PM, John Houghton wrote:
>
>
>
> > On Mar 29, 10:01 pm, Anne and Kevin Morrissey
> > <anneandke...@comcast.net> wrote:
> >> Is it possible to simply shoot too many frames? I shot twelve, and I
> >> might have gotten by with ten or eight.
>
> > 12 is overkill. Given that the horizontal fov would be around 94
> > degrees, 6 shots is normally considered adequate (36% overlap). Try
> > stitching the alternate images.
>
> > John
>
> > --
> > You received this message because you are subscribed to the Google
> > Groups "PTGui" group.
> > To post to this group, send email to pt...@googlegroups.com
> > To unsubscribe from this group, send email to ptgui
> > +unsub...@googlegroups.com
> > Please do not add attachments to your posts; instead you may upload
> > files at
> >http://groups.google.com/group/ptgui/files
> > For more options, visit this group athttp://groups.google.com/
Erik Krause
> The problems with 12 shots around may have been that because of the
> very large overlap, the control point generator is likely to generate
> points between non-neighboring images.
Of course I know this effect since the early days of panotools 10 years
ago. But I still wonder why this is the case from a theoretical point of
view. If I have images perfectly corrected for lens distortion without
any parallax errors image 1 and 3 should align equally well as images 1
and 2. A 360� panorama from all odd numbered images should be
geometrically the same like one from all even numbered images or all images.
I can only think of an effect in the optimization algorithm. Joost,
could you comment?
--
Erik Krause
http://www.erik-krause.de
John Houghton
> Am 30.03.2010 08:10, schrieb John Houghton:
>
> Of course I know this effect since the early days of panotools 10 years
> ago. But I still wonder why this is the case from a theoretical point of
> view. If I have images perfectly corrected for lens distortion without
> any parallax errors image 1 and 3 should align equally well as images 1
> geometrically the same like one from all even numbered images or all images.
I agree that large overlaps should not matter, and may not for
rectilinear images. However, ss you know, you cannot eliminate
parallax when using a fisheye lens because the apparent entrance pupil
position varies with the angle of incidence of the light rays. The
optimum NPP position is therefore chosen to suit the angle at which
the seams will be placed. Images 1 & 3 will join at a different angle
to images 1 & 2, so the NPP will not be ideal for both.
Another consideration is that many people may well overlook the
possibility of there being control points between non-neighboring
images. I know I have fallen into this trap myself occasionally.
Badly placed points don't always get deleted when using the "Delete
worst points", and if they are semi hidden in this way, resulting bad
alignment can be very puzzling. The control points table reveals all,
of course.
John
Anne and Kevin Morrissey
I have done some additional testing and found that eight images seems
sufficient (at least in my living room)! But because of the large
FOV of the 15mm lens, the front/back standard of the head was visible
at the bottom of the frame.
Heretofore I had dealt with this problem in an extemely inaccurate
manner (by changing the angle at which the camera was mounted on the
head itself, so as to "swing" the lens out of the way)! Obviously,
this changed the position of the entrance pupil from its intended
placement by Bogen!
Much easier, I've found, and more accurate, to simply tilt the camera
body up a bit so the image in the viewfinder is above the projected
image of the head, and mounting it properly in parallel with the
front/back standard. I can say fairly definitively that it was too
much overlap.
And, of course, thanks to you both, again -- so much -- for your
detailed, prompt and astute observations. It's a pleasure to use PT
Gui!
Kevin Morrissey
On Mar 30, 2010, at 11:06 AM, Erik Krause wrote:
> Am 30.03.2010 08:10, schrieb John Houghton:
>
>> The problems with 12 shots around may have been that because of the
>> very large overlap, the control point generator is likely to generate
>> points between non-neighboring images.
>
> Of course I know this effect since the early days of panotools 10
> years ago. But I still wonder why this is the case from a
> theoretical point of view. If I have images perfectly corrected for
> lens distortion without any parallax errors image 1 and 3 should
> align equally well as images 1 and 2. A 360° panorama from all odd
> numbered images should be geometrically the same like one from all
> even numbered images or all images.
>
> I can only think of an effect in the optimization algorithm. Joost,
> could you comment?
>
> --
> Erik Krause
> http://www.erik-krause.de
>
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui" group.
> To post to this group, send email to pt...@googlegroups.com
> To unsubscribe from this group, send email to ptgui
> +unsub...@googlegroups.com
> Please do not add attachments to your posts; instead you may upload
> files at
> http://groups.google.com/group/ptgui/files
> For more options, visit this group at http://groups.google.com/
mick crane
Anne and Kevin Morrissey wrote:
>
> I have done some additional testing and found that eight images seems
> sufficient (at least in my living room)! But because of the large
> FOV of the 15mm lens, the front/back standard of the head was visible
> at the bottom of the frame.
you can do this in ptgui itself but I forgot the exact procedure, I
think you need to know the field of view first . someone will know.
Erik Krause
> I agree that large overlaps should not matter, and may not for
> rectilinear images. However, ss you know, you cannot eliminate
> parallax when using a fisheye lens because the apparent entrance pupil
> position varies with the angle of incidence of the light rays. The
> optimum NPP position is therefore chosen to suit the angle at which
> to images 1& 2, so the NPP will not be ideal for both.
Agreed. But if all objects in a panorama are more or less far away there
are no parallax errors. Given I shoot a landscape spherical using a
fisheye from a pole, the parallax error should be far less than the
usual stitching error between non-neighboring images.
I suspect lens distortion being not perfectly rotational symmetric. Some
years ago I had a discussion with Stephan Stoske. He distributes a
program which determines and corrects lens distortion matrix based
(german only): http://www.stoske.de/digicam/Artikel/verzeichnung.html
If you scroll down the page almost to bottom you can see two greyscale
images visualizing the amount of distortion of a real lens measured by
his program. If you look at this the approximation of panotools and
PTGui is pretty bad.
However, this might be only one part of the problem. The other one might
be the optimizer, which gets confused by the two-step feedback. But
that's something Joost should comment.
Eric O'Brien
In the image you uploaded there are stitching errors on areas of far
walls that appear to be at essentially the same distance from the
camera. That is, the location where I see a bad seam is essentially
at a single distance from the camera on a relatively distant wall.
Those particular errors are not going to be caused by a misaligned
camera (that resulted in parallax effects). What this looks like to
me is that the lens optimization values are WAY off.
If you're using the Project Assistant, I suggest you try *bypassing"
that, switching the interface to "Advanced" and going through the
process manually. FYI: automatically generated control points tend to
clump near the center of fisheye images. To get the best lens
optimization values I have found it necessary to manually add some
control point pairs out toward the edges of the images.
eo
>> To unsubscribe from this group, send email to ptgui+un...@googlegroups.com