PDA

View Full Version : D300 + Nikon 10.5 fisheye



Ion
09-01-2008, 09:23 AM
Hi all,

I have been experimenting with my brand new NN5L at home, and it seems that I cannot get the rail settings right, I've used those from the NN website, but there remain parallax/stitching errors (used PTGui).
Anybody who has the same camera/lens and the same problem, or maybe does not have the problem...

I have added a sample picture. Camera was tilted -10 degrees for smaller nadir footprint.

Like to here your opinion.
Thanks and have a nice day!
Wim.

Ion
09-01-2008, 11:41 AM
Here you can see my lower rail settings when the 'cross' is in the middle. 63-64 mm...
Line were accurately drawn from image corners (and then cropped).

John Houghton
09-01-2008, 02:19 PM
I have two suggestions:

1) The settings from the Nodal Ninja site may or may not be reliable. Verify the upper rail setting by specifically checking for parallax, as described in http://www.johnhpanos.com/epcalib.htm.
2) Shoot a test panorama in a reasonably large venue that has plenty of nice features for control points in all directions. In your panorama at home, the canvas above is very revealing of stitching errors, yet it's not easy to find matching features beween adjacent images for assigning control points in that large area. (Try assigning points there manually yourself). Moreover - being flexible, the canvas may flap about in the wind between shots.

When you have done a successful stitch, you can then use the saved project file as a template, which can be applied to future similar sets of images to greatly help the stitching process when control points might be scarce as in this one.

John

MrB
09-02-2008, 06:20 AM
John,

I use exactly the same setup as you (D300 / 10.5 / Nodal Ninja 5L). I think the problem that you are seeing could be caused by the camera sensor being off center. This would mean it would look like everything is centered but things would still be off. These are the setting that I use. Lower rail 59.5mm (well somewhere between 59mm and 60mm). Upper rail 81mm. When I shoot I also use a -15 degree tilt. If you upgrade to the full Nodal Ninja 5 then this angle will have a stop to lock at that position too.

A HDR example of this setup in action is here http://www.bilsland.co.uk/greatwitleychurch/

I hope this sorts things out for you. If anything isn't clear enough let us know.

Bob Bilsland.

Ion
09-02-2008, 10:06 AM
Hi Bob,

Thanks for the reply!
I hope it's not my camera's sensor that is off centre... will be spending an awful lot of thime then finding the entrance pupil settings.
Do you shoot your zenith at 90degrees? I have made another pano just now and it seems that the parallax is mostly (if not only) apparent on the top and bottom of the pano (I do not shoot the nadir).

I use a -10 degree tilt, otherwise the tripod legs are in view.
Thank you all!
Wim.



John,

I use exactly the same setup as you (D300 / 10.5 / Nodal Ninja 5L). I think the problem that you are seeing could be caused by the camera sensor being off center. This would mean it would look like everything is centered but things would still be off. These are the setting that I use. Lower rail 59.5mm (well somewhere between 59mm and 60mm). Upper rail 81mm. When I shoot I also use a -15 degree tilt. If you upgrade to the full Nodal Ninja 5 then this angle will have a stop to lock at that position too.

A HDR example of this setup in action is here http://www.bilsland.co.uk/greatwitleychurch/

I hope this sorts things out for you. If anything isn't clear enough let us know.

Bob Bilsland.

John Houghton
09-02-2008, 10:39 AM
Don't automatically assume that stitching errors are due to parallax errors. They are frequently due to other causes - e.g. inadequate optimization and poor control point placement and distribution. Note that an offset of the lens axis with respect to the centre of the sensor is quite common. PTGui can make allowance for this if the lens shift parameters (d & e) are included in the optimization process. This is always recommended when a fisheye lens is used.

John

MrB
09-02-2008, 12:01 PM
Wim,

Yes a sensor offset is nothing to worry about. Just think how many pixels they fit into the size of the sensor. The D300 sensor is only 23.6mm x 15.8mm and in that size Nikon fit 4288 pixels x 2848 pixels. Now just think if that sensor is just 0.1mm out that equates to just under 20 pixels!

When I shoot my panoramas I shoot 6 photos round tilted 15 degrees down and a single zenith shot at 90 degrees up. I shoot at 15 degrees down because I don't open my tripod out fully, I have the legs spread as little as possible, just enough to be stable. This combination produces a very small hole to patch below.

Bob Bilsland.

Ion
09-02-2008, 12:30 PM
Dear Bob,

Ok?©, an off-center sensor sounded like some kind of bad camera-disease, but I guess it's not that big of a disaster, ??f it is the case.
I do the -10/-15 degrees trick as well, it's saves you a whole lot of photoshop work or it just eliminates the nadir shot if a nadir cap is used, thanks.

I have stitched my photo's now with Realviz and I have to say I get far better results that with PTGui... not p?©rfect, but close to, so I'm getting confidence it's software-dependent (well... it's depending on my skills configuring the software I guess :-) ). The problem persists so maybe I have to tweak the panohead some more too.

About the zenith shot, I have now tried and angle of 60 degrees instead of 90 and the results are great. When you stitch the images without the zenith shot, the zenith hole is relatively small, so not that much info was gathered from the zenith shot. A 60 degrees photo gives you far more control points with the other photo's, certainly when photographing a while ceiling or so.

I'll keep on trying...
Have a fine day!
Wim.

John Houghton
09-02-2008, 01:33 PM
You should be abe to get comparable results with PTGui as you have done with Realviz. If you care to make a set of images available somewhere, I would be happy to try a stitch with PTGui and see what might be possible if that would be of interest.

John

Ion
09-02-2008, 02:51 PM
Dear John,

Thanks a lot, that would be of great help to me!
I have uploaded the files I have been working on last, I have also included the PTGui and Realviz stitch. I have used the WIM5275 +60 degrees zenith shot, the others are for demonstration.
BTW: I have changed my NN5L back to the recommended setting from NN and this is the result.

Thanks again, I'm quite curious what you have to tell me!
Greetings, Wim.

Oh yes, the files are here: http://users.pandora.be/sijoschooltoneel2007/

John Houghton
09-02-2008, 05:33 PM
Wim, I stitched the images and put the result and project file together in a zip archive at www.johnhpanos.com/ion.zip (http://www.johnhpanos.com/ion.zip). There is some horizontal parallax visible in the images and rather more vertical parallax, so you might like to check the head adjustment specifically for that. That may account for the not very sparkling optimization report, though it's late and I haven't spent much time investigating control points. I did place a few points manually in the expanses of blank wall.

John

Ion
09-03-2008, 02:29 AM
Hey John,

Thanks a lot for helping me out!!!
The vertical parallax, you mean I have to change the lower rail setting or am I mistaken? That would make sence since it's that one I had 'centered' on 63mm...

I wonder what happens if you buy an 'Absolute' panohead by 360precision... it's precise yes, but what if your camera is off.

Wim.

John Houghton
09-03-2008, 04:01 AM
By vertical parallax, I mean image shift when the camera is roated about the horizontal axis, as when you rotate the camera upwards to take the zenith shot. This is taken from your panorama, and shows the parallax effect between the zenith shot and a horizontal shot:

http://www.johnhpanos.com/ionplx.gif

However, it's virtually impossible to eliminate parallax completely when using fisheyes because the entrance pupil is not a single fixed point in the lens. Its position varies according the direction of the incident light rays. For light rays entering the lens at extreme angles like 90 degrees to the lens axis, the entrance pupil is close to the front surface of the lens. OTOH, for light rays entering directly towards the front of the lens, the entrance pupil is much further back into the lens, maybe 10-15mm. So as the light rays shift around from 0 to 90 degrees, the entrance pupil gradually moves forwards.

When you set up the pano head to avoid parallax, you therefore adjust the camera position to suit the angular position where the seam will be located. So for 6 shots around, the yaw increment is 60 degrees and centre of the overlap between two adjacent images will be at 30 degrees to the lens axis. So to eliminate parallax at the image join, you arrange for the camera to rotate about the position that the entrance pupil is in for 30 degree incident rays.

The problem with the your zenith+60 image is that the corner of it overlaps with the central part of a horizontal image. Consequently, the entrance pupil is not in the same place for each image at the join, so parallax is inevitable. There are further ramifications at http://www.panoramas.dk/panorama/fisheye-NPP/parallax.html.

Parallax is only a problem with near objects in the overlap regions. It's best to avoid putting control points in such places if there is risk of significant parallax effects. The automatic cp generators won't avoid doing this, so it's best to be on your guard.

Another problem might be that the camera is not mounted on the upper rail such that the lens axis is accurately parallel to the rail. A small amount of camera tilt with respect to the rail will swing the entrance pupil away from the horizontal rotation axis at the end of the top rail.

The 360Precision head cannot eliminate the inherent parallax problems with fisheye lenses, but if you can manage to successfully stitch with templates and without control points, then you don't have to worry about where control points are placed.

John

Ion
09-03-2008, 05:21 AM
Thanks for this info. It's quite hard figuring this all out.

>Another problem might be that the camera is not mounted on the upper rail such that the lens axis is accurately parallel to the rail. A small amount of camera tilt with respect to the rail will swing the entrance pupil away from the horizontal rotation axis at the end of the top rail.

=> Well, I have experienced this too. Actually, when you untighten the vertical arm (the lower know, to change lower rail settings) you can even 'twist' the vertical arm a little bit (around it's vertical axis) on my NN. It's really not that much but enough to swing the entrance pupil out of place like you said. So not only the lower rail setting is important but also how you tighten it, you have to make sure to press the arm firmy against the side of the rail to be sure it's parallel.
I hope you understand my explanation, I'm trying to be as precise as possible.

Wim.

John Houghton
09-03-2008, 05:53 AM
Wim, I don't think you need to worry too much about the setup of your pano head. Considering how close the camera is to the nearest features, the panorama has come out very well. In larger venues and outdoors, you shouldn't have any problems. One odd thing: while I noted that the images were all given the same exposure, there are quite big variations in the brightness of the images. If they were generated from raw originals, perhaps the conversion settings weren't constant or some were set to auto.

John

Ion
09-03-2008, 06:29 AM
The exposure really are the same. Only whitebalance was changed in camera raw. Perhaps it's because of the very unregular illumination in my room (it's a small but long room and there are spotlights in it pointing at my desk, leaving the rest of the room dark. It was also becoming dark when I took the pics, so no daylight illumination was apparent.

Just one more thing...
According to this table the entrance pupil of the nikkor 10.5 fisheye is located at +/- 46mm from the lens base. If I measured that correctly it should be in front of the golden circle, leaving the upper rail setting at about 90mm...
Do you have any idea what the distance change is of the entrance pupil of a (this) fisheye when the light is coming in straight or at a 90 (well almost) degrees angle?

I have read the entrance pupil changes with the change of the aperture as well, do you know if this is true/has any notable influence?

Thank you again for all the great advice!
Greetings, Wim.

John Houghton
09-03-2008, 08:05 AM
The position of the entrance pupil for rays at 90 degrees to the lens axis can be seen just by looking at it, like this:

http://www.johnhpanos.com/nik10-ep.jpg

Likewise - by looking, I would estimate the on-axis position to be roughly 3mm or so behind the gold band. I haven't measured it.

The entrance pupil IS the aperture, or rather the virtual image of it formed by the lens elements in front of it. If you open up the aperture, something else might then limit the effective aperture, and that then becomes the entrance pupil. With the lens wide open, the imaging of near objects can become fuzzy and out-of-focus. With no sharp edges, the parallax effects become rather diffuse and less of a problem.

John

Ion
09-03-2008, 08:30 AM
John,

Valuable info again...
I really think you should write a book or so on this topic, you seem like a walking pano-encyclopedia.
You surely would sell at least one book... to me :-)

Greetz, Wim.

Ion
09-03-2008, 12:00 PM
Hi John,

I've looked again at the stitch you've made from my images... I cannot find any stitching errrors...
It's quite amazing you got this all together. Did you have to do many handwork or has this been done (semi)automatically by PTGui?

I have been experimenting with the optimizer, the results are getting better, but not as perfect as your stitch.

And... lens parameters, are they the same for all lenses (well all of the same type, like my 10.5 nikkor) or does that change for each one?
If it's the same it would be a good idea to make some sort of database for those parameters.

Greetz, Wim.

Ion
09-03-2008, 12:43 PM
You hadn't had a chance to react on the previous post and here I am again... :-) I should be sending you a bunch of flowers or so to thank you, but I suppose you live somewhere on the other half of the planet... :-)


>When you set up the pano head to avoid parallax, you therefore adjust the camera position to suit the angular position where the seam will be located. So for 6 shots around, the yaw increment is 60 degrees and centre of the overlap between two adjacent images will be at 30 degrees to the lens axis. So to eliminate parallax at the image join, you arrange for the camera to rotate about the position that the entrance pupil is in for 30 degree incident rays.

=> Can you try to rephrase this for me? Seems like inportant info and I don't really get it.
Thanks!

John Houghton
09-03-2008, 01:15 PM
I was reacting as you posted the last item:

The stitch I produced was directly out of PTGui without any postprocessing, but was not a completely automatic stitch. There was intelligent manual intervention to improve the chances of a good output. The automatic control point generator ofen doesn't place control points with an ideal distribution along the overlap areas. It has difficulties matching up the images in the more distorted areas near the corners. A good spread of control points is needed to enable the optimizer to accurately correct the lens distortions. It can only work out appropriate lens parameters that align the images in the corners if there are control points there. So it's a good idea to do a quick check to see that the control points look reasonably ok if stitching as not gone perfectly well. If not, you can place extra control points in areas where there are alignment (stitching) errors and hope that the optimizer will coax those areas into better alignment. Of course, if you have evaluated good lens parameters in a previous project, these can be used in future projects and can be applied conveniently with a template file. The optimizer doesn't need to work them out all over again and so a well distribued set of control points is not then so important.

I imagine that lens parameters should be the same for each model of lens, but they will vary according to the sensor size in the particular camera the lens is used on. Also, a crop applied to the image (as is commonly done with circular fisheyes), will also change the lens parameters. The shift parameters d and e need to be optimized as they are not constant - probably not even for the same lens on the same camera for different shoots.

Panoramas taken in small rooms with more-or-less blank walls and ceilings are not easy to stitch perfectly. Some people create artificial features for control points by sticking bits of tape on the walls (to be cloned out of the final image). You should have an easier time with larger venues and outdoor scenes, and with nothing very close to the camera, parallax effects will be largely eliminated.


As to the entrance pupil clarification requested:

If you shoot a panorama with 6 shots around, the yaw increments by 60 degrees for each shot. It follows that the seams where the images join are every 60 degrees too (there being 6 in 360 degrees). So if you imagine facing straight ahead at yaw position 0 for the first shot, the two joins to the adjacent images will be made at 30 degrees to the left and 30 degrees to the right, i.e. at 30 degrees to the lens axis pointing straight ahead. It's at this part of the image that we don't want to see any parallax effects that will interfere with the image blending and prevent a seamless join. Given that the entrance pupil moves forwards as we move away from the lens axis, it's the entrance pupil position corresponding to the 30 degree direction that is the point about which the camera should be rotated. The entrance pupil will then be in the same position for each overlapping shot, and features in the vicinity of the seam will not be affected by parallax.

To look at it (literally) another way, if you position yourself in the panoramic scene precisely at a point where one of the seams will eventually be placed when the images are stitched, e.g. 30 degrees off-axis in the first shot, then you need to see the entrance pupil in the lens occupy exactly the same position for the first shot and also for the next shot (you will be on the right of the first shot and the left of the second shot). If the entrance pupil is in the same place for the two shots, then parallax will be eliminated at that point where the images join. This is the basis for my method of setting up the pano head correctly, which I referred to earlier in http://www.johnhpanos.com/epcalib.htm. It only takes 10 minutes or so to set up the head this way.

John

pf127167
09-03-2008, 02:18 PM
Wow guys I hate to interrupt a very interesting conversation but John I am almost a complete novice and you sound like someone that can give a few quick pointers on the proper workflow to create HDR panoramas.

Should I shoot RAW?
Do I mege to HDR first or do I stitch the pano first?
etc, etc.

Again, I apologize for butting in and thanks in advance for any pointers you might be able to give me.

Jeff

John Houghton
09-04-2008, 12:46 AM
Jeff, Shooting RAW is the preferred option. It gives you the maximum flexibility and quality. You can easily make corrections for chromatic aberration, white balance and vignetting, depending on your RAW converter. If speed of shooting is important, it can sometimes be better to use JPEG to avoid hold ups due to buffer filling up on write-outs to the memory card.

For HDR, merging images before stitching is the simplest technique. Sitching and then merging enables you to preview the entire panorama when tweaking the parameters, which should be an advantage. One disadvantage is that the blender may not place the seams in exactly the same place for each stitch, and this can occasionally lead to mismatches. You should do your own tests and see what works best for you.

John

Ion
09-04-2008, 02:37 AM
Hey Jeff,

As you may have read, I'm not the expert here, but I have had good results with the HDR function of PTGui. Just shoot your brackets in RAW, convert them (make sure whitebalance, CA is ok) to jpeg and open them all together in PTGui. If you have shot them from a tripod (I think you will have) you best 'link' the branckets when PTGui asks you. That means that PTGui does not try to align the bracketed images but merges them together like they are. You really need the image brackets to be perfectly aligned for that, but since you were using a tripod that should be the case.

After that workflow is quite the same as a normal image I suppose. I do have better stitching results using HDR than I have with single shots. How placing of control points by hand is done with all those images I don't know, John any idea?

See you!
Wim.

Ion
09-04-2008, 03:12 AM
Oh, John, I get it it now!
Lot's of the fisheye entrance pupil confusion has been cleared out here. Thanks.

Ion
09-04-2008, 04:36 AM
I am breaking my head here over this Nodal Ninja...

I'm taking pics, turning 'em around in photoshop, comparing, adjusting, agian again... doesn't seem to be working to get the lower rail setting good.

Two problems: I keep getting left/right shift (so perpendicular to the lower rail), but I have checked evenrything over and over: camera is perfectly straight on it's mounting plate, that one is perfectly straight on the upper rail. I set that one on exactly 90 degrees etc... and that shift will not go away.

Second thing I have noticed is that my nodal ninja does not turn around (the cross on) the knob perfectly... If I measure the distance from the vertical arm to the centre of the cross (yes the cross in perfectly centered on the know, I've measured it), then I rotate the NN to the different click stops I measure a difference of 0.5 to 1mm.

Another thing I am confused about is: if I take two nadir shots (one at 0 degrees, one at 180 degrees), and I turn one of them 180 degrees in photoshop, should the surroundings align well? The knob alligns quite oke (except for the small shift I talked about) but the surroundings are far off...
Or the other way around if I stitch those with PTGUI, I get fairly good results for the surroundings but the knob is visible twice (and also the small sawblade pattern, but that I from the shift if I have read your pages well John).
Have included that pic.

I'm really freaking out here ;-) somebody help! :-)

Ion
09-04-2008, 04:39 AM
This one is the same as above but from 6 shots and stitched...

John Houghton
09-04-2008, 09:20 AM
Wim, The first of your sample images shows parallax shift of the knob with respect to the background. You therefore need to move the vertical arm inwards towards the knob by 3-4mm. Also, you need to rotate the camera clockwise a smidgin (as viewed looking at the top of the camera i.e. the hot shoe).

John

nick fan
09-04-2008, 10:45 AM
hi wim,

the parallax shift is related to distance. the close the subjects, the more pronounced the shift. Since fisheye lens do not have fixed NPP and parallax is distance dependent, it is very difficult get a parallax free pano in a small crowded place. you should calibrate the pano head for scenes that you are going to take. Why not try easier ones such as a garden or a living room pano first? after you get some experience and enjoyment, you can come back to tackle the more challenging scene.


Nick

PS since the rotator tension knob is very close (relative to other subjects) to the lens, I suggest you to ignore it. You should be able to align the NPP using external references only. Use you current setting as a starting point. Align the NPP on the upper rail first, then the lower rail setting according to Manfrotto 303 SPH's manual. See page 5-7.
http://services.manfrotto.com/303SPH/content/303SPH_manual.pdf

Ion
09-04-2008, 12:32 PM
Hi guys,

=> Bringing vertical arm back 4mm: check.
(That brings me back to the original 59.5 mm ;-) )

=> Ignore the knob: check.

=> Turning the arm clockwise a smidgin (what a funny word :-) ): check.

Result: A stitch of the same small room with only one tiny little stitching error (perhaps with the hint from John on CP optimization, I can remove it)!
Think I will glue these settings, to never loose them again :-)))



On another forum Matt from 360precision said: "The D200/10.5 setting for six shots is 83mm and 86mm on the D300. Matt 360Precision.com" (http://www.panoguide.com/forums/qna/4486/)
I suppose he's referring to respectively the lower and upper rail settings? But without the CP I suppose. Does anyone know when the table of NN settings on the website has gone and when it will be back? Then I'll be able to check if these match.

See you and again thanks, looks like this is the busiest topic on the forum :-)

Best regards, Wim.

John Houghton
09-04-2008, 02:46 PM
On another forum Matt from 360precision said: "The D200/10.5 setting for six shots is 83mm and 86mm on the D300. Matt 360Precision.com" (http://www.panoguide.com/forums/qna/4486/)
I suppose he's referring to respectively the lower and upper rail settings?

Matt was not referring to lower rail settings at all. 83mm refers to the upper rail setting with the D200, and 86 refers to the upper rail setting with the D300.

It's of no importance what anybody else tells you the settings are. You can so easily check for parallax by taking some photos yourself and know for sure whether your settings eliminate parallax or not. If they don't, adjust them until they do. You then know that your settings are correct. Whether they are the same as anybody else's is of no particular interest.

But do try to get out and shoot a panorama in a much bigger venue where parallax is not so important an issue and see if you can manage a stitch without any errors.

John

Ion
09-05-2008, 03:09 AM
Okidoki thanks.

I was just thinking, if the 360precision guys make a panohead that cannot be changed and is qualified for making stitches out of templates, that there should exist "th?© settings" for the NN5 with my D300 and 10.5fish (well, of course for all the other camera/lens combo's as well). I like scientific ways of seeing things not always the trial and error mode and since this is quite a popular combinaton I guessed s??meone should know 'th?© settings' :-)
Guess I'll be trialanderroring a bit further :-) but I'm getting close now.

Have a nice day,
Wim.

Terrywoodenpic
09-05-2008, 04:20 AM
Okidoki thanks.

I was just thinking, if the 360precision guys make a panohead that cannot be changed and is qualified for making stitches out of templates, that there should exist "th?© settings" for the NN5 with my D300 and 10.5fish (well, of course for all the other camera/lens combo's as well). I like scientific ways of seeing things not always the trial and error mode and since this is quite a popular combinaton I guessed s??meone should know 'th?© settings' :-)
Guess I'll be trialanderroring a bit further :-) but I'm getting close now.

Have a nice day,
Wim.



It would be nice if it were true, that there was a scientifically exact setting for each and every camera and lens combination.
However the fact remains that manufacturing tolerances play their part and although the exterior of lenses and cameras tend to be very accurate and repeatable,
The same can not be said for lens settings and sensor placements.

There are small but significant differences in the settings achieved by most photographers to arrive at "Perfect Stitches"

This is all unfortunate but true... equally true is the fact that some seem able to set better control points than others, and some need to do less fine corrections in photoshop than others.

All very unscientific.... more of a " Dark Art" perhaps? I think John is some sort of master "Wizard" when it comes to stitching... but it is strange how luck improves in proportion to practice and understanding.

Ion
09-05-2008, 05:02 AM
Yes indeed it certainly is more then numbers and photographs...
But I keep wondering how the 'Absolute' panoheads work by 360Precision... if there are, for each lens or camera, differing settings cause of their interior, you sure must have some luck to have a camera or lens that matches the unchangable Absolute panohead. If any parallax occurs cause of a sensor shift or I don't know what, then you might as well buy a new one untill you get better luck.
Don't know if this makes sense, I'm just wondering...

Wim.

John Houghton
09-05-2008, 06:09 AM
A sensor shift cannot cause parallax because it has no impact on the entrance pupil position, but a new template file would be required that took took account of the shift.

John

Ion
09-05-2008, 06:43 AM
Thank you.

I have plenty of stuff to think about now.
You will hear from me again when I'm struggeling with the settings again ;-) (quite soon I expect)

Bye, Wim.

Ion
09-07-2008, 02:37 PM
Hello everybody,

I have intensively been busy panoram'ing and i would like to say that I have, because of your - en esp?©cially John's - help I have managed to do to several successfull, errorfree stitches with PTGui.
Some outdoors, some indoors, with close and far away objects. I'm feeling confident thanks to you.

Just wanted to say that.
Best regards,
Wim.

dume
06-23-2010, 10:47 AM
does anybody know about D300+nikkor 10.5mm setting wtih NN5??