Will it work? - NN 3 MkII + Canon 500D + Tokina 11-16 f2.8

[hindenhaag]

Hi Jens,

When you turned your camera to the right, and the reference point moves with the camera to the right as well, you have to move your camera forward on the upper rail. This is what I see from your pics. So you have to move forward from 107 > to 104 for example. When the reference moves to the opposite side, then you have to go backwards.

You have to set WB to a fixed setting like "sun", not to Auto. You might see differences in your pano taken with WB at "Auto"

Focus has to go to manual as well. Otherwise the camera will focus for each shot to something else. Set Focus to manual on your camera body. Google "Hyperfocal Distance", think it is dofmaster, and look it up for the chosen camera, lens and aperture. Only this will give you the best "Depth of Field" for you settings.

You may just use John's methods to test your NPP. I find the belgium grid method a little bit too complicated. "Keep it simple".

Regards

Heinz

[No-Nonsense]

Hello Heinz,

thank you for your help!

When you turned your camera to the right, and the reference point moves with the camera to the right as well, you have to move your camera forward on the upper rail. This is what I see from your pics. So you have to move forward from 107 > to 104 for example. When the reference moves to the opposite side, then you have to go backwards.

I fear I made a fault determining the direction I turned the camera. I did a new test yesterday with the following setup: 11mm, F11, focus at infinity, ISO 100, spot metering (to have the area of interest correctly illuminated).

In the second image set I started at the maximum of 107mm on the upper rail. My impression is that this set clearly shows that I have to set the upper rail beyond 107mm (actually 1-3 mm I would suspect).

For the first image set I did a trick to go beyond the 107mm. The screw on the rail is long enough so that I could lay some folded paper underneath the CP-1 plate to go beyond the limit stop.

Now I will try to get the NN T-Adapter. If I can then get a good setup with my gear, I will see if I find some shop that could shorten my CP-1 plate by ~2-3mm.

What do you think?

Thanks in advance, greetings,
Jens

[No-Nonsense]

Hello!

This is an update on my progress with the Canon 500D/T1i, the Tokina 11-16mm, and the Nodal Ninja 3:

In the meantime I bought a Nodal Ninja T-Adapter from the local distributor Thaler Design, which allows to move the camera on the upper rail beyond the 107mm that were possible with the CP-1 plate. With this I tried various NPP setup methods (German speaking people can read the full history in my thread in the Panorama-Community) and am convinced that the upper rail must be set to something between 108mm and 109mm.

With the best setup I did some test shots with the following shooting pattern: either three rows à 8 pictures at 0°, + 45°, and -45° or one row à 8 pictures at 0° and two rows à 6 pictures at +45° and -45°.

Unfortunately I was not able to get any good results using Autopano Pro. The stitched panoramas always suffer from some skips that I was not able to remove by manual optimization of CPs and links. I tried the various settings in Autopano Pro (number of CPs, optimization settings, lens distortion correction, ...) without success.

I was about to give up when I gave the current trial of PTgui a try. PTgui was able to stitch my images without any issues when selecting the option "Minimize lens distortion: Heavy + lens shift" without any manual intervention.

You can find some examples of both my Autopano Pro and PTgui results in my Picasa gallery.


I have started a thread regarding my issues with Autopano Pro in the Autopano forum some days ago, but unfortunately nobody has answered to it as of now. I would like to save money and therefore don't want to spend money on PTgui, as I already own Autopano Pro.


Does someone on this forum has any ideas what I could try to get good results with Autopano Pro? I have uploaded two sample images sets (~100MB each set) to my Dropbox:
http://dl.dropbox.com/u/944445/2010-08-02.7z
http://dl.dropbox.com/u/944445/2010-08-12.7z


Thanks in advance, kind regards,
Jens

[hansnyberg]

They stitch perfect in PTGui but I can see you have a very large horizontal shift. 55 pixels shift for this size of images is really large and it might be that autopano has problems handling this. I have no experience with autopano but in PTGui this is taken care of by optimizing for the shift.

As far as I can see no large parallax but the largest controlpoints errors come from upper and lower row connection to the centre row. That might point out that it is slightly wrong but it may also mean that your camera is not parallel to the axis of the spherical arm. Something that the T-adapter could be responsible for.
However the room does not give you much options for good controlpoints between the rows either.

Hans

[No-Nonsense]

Hello Hans!

Thank you for experimenting with my test image sets and taking time for a reply!

They stitch perfect in PTGui but I can see you have a very large horizontal shift. 55 pixels shift for this size of images is really large and it might be that autopano has problems handling this. I have no experience with autopano but in PTGui this is taken care of by optimizing for the shift.

Could you please go a little bit more into detail why you consider the horizontal offset as very large?

My understanding of the lens shift parameters is that they specify that the offset from the center of the distortion of both camera and lens. Is this correct? Or could the large values also be caused by mounting the camera off center on the Nodal Ninja?

I also have noticed that the lens shift parameters vary very much between different image sets I have taken so far (with different lower/upper rail setups). Could this be due to different setup? How can I ensure that I mount the CP-1 or T-Adapter really right to the camera and then also to the rail (there is some tolerance). Do I really have to take care of the camera body not beeing 100% plain (I have read here to fix this with tape)?

As far as I can see no large parallax but the largest controlpoints errors come from upper and lower row connection to the centre row. That might point out that it is slightly wrong but it may also mean that your camera is not parallel to the axis of the spherical arm. Something that the T-adapter could be responsible for.
However the room does not give you much options for good controlpoints between the rows either.

What I found impressing is that PTgui was even able to stitch all my previous image sets taken where I used the CP-1 and a small table tripod. To me it looks like Autopano Pro had really big issues with ultra wide angle lenses.


Could you please have a look at this:

Besides other NPP-setup methods I followed these two methods and combined them:

1. On the Panorama-Community I read about a method to setup the lower rail by using exactly the same method as used for the upper rail: aligning something near (e.g. a small arrow attached to the tripod) with something far (e.g. the ground) with the camera facing down and while turning the Nodal Ninja 180°.
I have two images that show my calibration result: image 1 and image 2.
What bothers me a bit is that the pictures also show that the camera + lens are mounted off center (i.e. the turning point of the Nodal Ninja rotator is not in the image center with this setup method. On the other hand this is the only setup where I have nearly no parallax while turning by 180°.

What do you think? Should I try with the rotator mounting screw in the image center instead?

2. I tried "The Grid" method described here: Finding the "Point" and "La Grille" - "The Grid". I used the lower rail setup from the other method, as I found calibrating both the lower and upper rail at once not very easy to do.
I have uploaded my calibration result images to this gallery.
With Autopano Pro I get a really bad stitching result. With PTgui it is much better, but still not perfect.
Now what I find strange is this: I see some kind of parallax that is only visible to the border of the images. It can be seen quite easily in the images 4 to 6: In image 4 and 5 the distance between the grid and the top edge of the rack are identical, but the distance between the grid and the rack-box below are not. Now comparing the distance between the grid and this rack-box in images 5 and 6 shows that there it is identical.

Could it be that PTgui crops my images (as Autopano Pro is only able to do with images taken with fisheye lenses) so that this parallax error does not have such a big impact in PTgui?


Thanks in advance, kind regards,
Jens

[hindenhaag]

Jens,

I tested your set 8/6/6 in autopano but I am getting crazy within 1.5 hours. I deleted all bad CP, lined it up, added manually, played around with settings, there is a lack of info in the upper and lower row. When you open up the CP control pic you see a lot of orange squares with bad CP. I deleted them to good conditions and tried to add CP manually. Most of the problems arise from the blue painting to the door and further on. With 6 around this is not enough overlap for autopano.

So I now downloaded the 8/8/8 set and tested it in PTGui. Created a .mov and I see a hacksaw cutting to the right. This means you should move your cam 1mm to the right in the LRS and try again.

I now will test this set in autopano, let's see what happens.

BTW: the shift is corrected nicely by autopano by using the "adapt Horizon Button". You can clearly see that this corrects stitching errors on the blue picture and on the upper part of the door frame for example.

Jens, benutze einfach ein Geodreieck um den rechten Winkel fuer CP-1 zu pruefen, und ein Lineal um zu checken, das der Adaptor mit der upper rail in der Flucht sitzt, d.h. das dein adaptor nicht schief sitzt. Du musst nicht duennes Tesa auf den Boden der camera kleben, alles Quatsch. Die Software korrigiert das. Die Leute, die das propagiert haben benutzen es wahrscheinlich selbst nicht mehr. Auf der einen Seite sind Sie super korrekt, auf dem naechsten thread erklaeren Sie, es kommt auf 1 oder 2mm nicht so sehr an.

Keep cool man!

Heinz

[hansnyberg]

Jens,


So I now downloaded the 8/8/8 set and tested it in PTGui. Created a .mov and I see a hacksaw cutting to the right. This means you should move your cam 1mm to the right in the LRS and try again.

Heinz

???
I see no hacksaw. The LRS is perfect.
http://www.panoramas.dk/test/jens.jpg

Hans

[No-Nonsense]

Hans,

thank you for dedicating this much time to my problem! I am really overwhelmed by the kind of help I receive here. This is really a great community!

I have also seen the slight hacksaw in the images stitched with PTgui. But when I would correct the lower rail away from the center (to the right?), then there would be parallax with the method I used to setup the lower rail. This is what confuses me. I will nevertheless try that tomorrow - maybe I am too focused on these calibration methods than on the final result. :)

Does "Adapt Horizon" really have to do with lens shift? I thought that lens shift is just another parameter of the lens distortion correction. In Autopano Pro I can only select to correct lens distortion and heavy lens distortion, but there is no option like in PTgui to handle lens shift. Maybe Autopano Pro already handles lens shift with the lens distortion options, but then not very well?

What settings did you use to process the 8/8/8 image set in Autopano Pro? I tried various settings and did not get really good results. I even tried to manually set all CP clouds, but that did not make it better (most of the time it made it worse).

Thanks in advance, kind regards,
Jens

[hindenhaag]

{post deleted}
Moderator note: Input is greatly appreciated in this forum and highly valued. We do need to try to keep on point with subject matter inside a thread and would encourage folks with tendencies to drift outside the subject matter to please simply start a new thread.
Thank you
Bill

[hindenhaag]

Jens,

I tried your second set. Without playing around with my preferences, it does not look bad. I think we should compare our preference settings. May be we should discuss this on our own. Not anybody might be interested in this. I can record my workflow and send it to you.

I am personally interested to set up a comparison PTGui/AutoPano to present this to our "normal users" to help them to set up basics to compare the software in reality with real pics. And I really like to test myself to prefer PTGui instead of Autopano. Actually I bought it because of the possibility to use different focal length in one pano set up as it is presented by the developers, lets us say 50mm at 0° and a different focal length in other rows. But actually a lot happens in Autopano without asking me and it is resulting in a "rare pano" without my understanding why this happened. Different to PTGui.

I am gonna sleep now:em0200:

Regards,

Heinz

[hansnyberg]

Hans,

thank you for dedicating this much time to my problem! I am really overwhelmed by the kind of help I receive here. This is really a great community!

I have also seen the slight hacksaw in the images stitched with PTgui. But when I would correct the lower rail away from the center (to the right?), then there would be parallax with the method I used to setup the lower rail. This is what confuses me. I will nevertheless try that tomorrow - maybe I am too focused on these calibration methods than on the final result. :)

Does "Adapt Horizon" really have to do with lens shift? I thought that lens shift is just another parameter of the lens distortion correction. In Autopano Pro I can only select to correct lens distortion and heavy lens distortion, but there is no option like in PTgui to handle lens shift. Maybe Autopano Pro already handles lens shift with the lens distortion options, but then not very well?

What settings did you use to process the 8/8/8 image set in Autopano Pro? I tried various settings and did not get really good results. I even tried to manually set all CP clouds, but that did not make it better (most of the time it made it worse).

Thanks in advance, kind regards,
Jens

Jens
As you can see from my image posted above there is no hacksaw when the images are processed and optimized correctly.
You may get hacksaw even with perfect settings if you do not optimize for shift.

The normal reason for shift is that your sensor is not aligned precise. Remember tat we are talking about extremely small dimensions . When the sensor is misaligned horisontaly with the lens the warping will be different in left and right side which has to be blended. Same for vertical shift. This used to be much less than your camera has. Normally I do not see more than 40 pixels shift.
However not only the sensor has shift. Also the lens can be misaligned, my Tokina 10-17mm gives me a difference of 25 pixels compared to my Canon 15mm. If the shift of your lens is the same way as your sensor shift they might ad together for example sensor 25pixels+ lens 25 pixels.

Hans

[No-Nonsense]

Hans,

thank you for your reply and the explanation of the sensor shift issue!

Jens
As you can see from my image posted above there is no hacksaw when the images are processed and optimized correctly.
You may get hacksaw even with perfect settings if you do not optimize for shift.

I think we are discussing different pictures. I provided two image sets. One of my workroom at home and one of my balcony.
The one you showed is my workroom and you are absolutely right, there is no hacksaw. The one the other Hans mentioned is my balcony and there is a slight hacksaw.

It is possible that I used different settings (lower and upper rail) for both image sets, as they were taken on different days and I experimented a lot. Sorry for that.

Greetings,
Jens

[hansnyberg]

Hans,

thank you for your reply and the explanation of the sensor shift issue!
I think we are discussing different pictures. I provided two image sets. One of my workroom at home and one of my balcony.
The one you showed is my workroom and you are absolutely right, there is no hacksaw. The one the other Hans mentioned is my balcony and there is a slight hacksaw.

It is possible that I used different settings (lower and upper rail) for both image sets, as they were taken on different days and I experimented a lot. Sorry for that.

Greetings,
Jens

Yes you are right. As you can see I made the screenshot with the pano in the background so that you could see it but Heinz seems to be more interested in harassing me than actually look at what I am showing.
The problem with the room is that you have very little features in the ceiling and some of the images has no features for controlpoints at all.

I had a look at you second set and yes there is a small hacksaw corresponding to 1mm LRS error. However as I pointed out several times +-1mm does not give problems even in very tight environments.
And this pano also stitches perfect however there is a lot of work actually because of the rotator in the images.
The rotator generates controlpoints and they give you errors in thousands of pixels.

After deleting them and also some bad ones in the sky I get a perfect stitch.
Here is a Pano and project file.
http://www.panoramas.dk/test/jens2.zip

I never include the rotator in the pano also because that will give you problems if you want to stitch the nadir in PTGui using viewpoint. Even with just a small part of the rotator at the bottom it may give you a shadow in the blending which needs extra work.

Actually I would say that the best way of shooting with your lens is 8 at -30 8 at +30 and 1 zenith at +90.
That may of course sometimes give you problems with no features in the zenith but in most cases it is easy fixed if you learn to take the zenith the same way always.

Hans
www.panoramas.dk

[hansnyberg]

Jens

If you compare image 5-6-7-14 -15 in the project file you can see that there is no parallax which will give you any problems even if the rotator shows a small hacksaw.

Hans

[hindenhaag]

Jens,

which version of autopano pro are you using. This is important to know, if it is possible to add a single CP to get better results in difficult situations. Yesterday I tried your second set in autopano giga and after deleting a lot of bad CP, link by link and pics by pics, I ended with a fair result, some stitching errors left. I do not work too much with autopano because I do not like the way CPs are placed and they way you have to del bad ones.

Just discussed this problem with Vincen, SkiVR. He told me, that there are more possibilities in the latest versions of autopano, even to add a single cp manually. I will have to try this.

So there is a possibility to get a good result with autopano pro as well, but I prefer PTGui, specially the latest beta version with a lot of new functions.

Uebrigens,

Heinz...