My Autopano Pro forum

The max size is a confirmed bug. I just forgot to adapt this code to fisheye lenses.

AlexandreJ wrote:

Doesn't seem to be good in ptgui either. Seems that pictures have a head issue.

Hmm, your ptgui preview window doesn't look too good at all. I deliberately left out the nadir, since it was handheld and doesn't align with the other images too well. I can't remember exactly what else I did in ptgui - I usually optimize afterwards for the parameters D and E which seems to have a significant effect on the alignment accuracy and delete some poorly placed control points. The point is that I could get a good stitch in ptgui, but not in APP. Perhaps it is user error...

I'm going to stand by my assertion that the panohead is set up correctly though - because I can get a good quality stitch from that image set (without handheld nadir) without using smartblend or retouching in photoshop, aswell as the hundreds of other panoramas I've managed to stitch successfully since acquiring my fisheye lens using the same head settings.  I also seem to have arrived at the same settings as several other panoguide forums users with the same 350d-sigma8-nodal ninja setup

Edit: N.B. I tilted the head up 5 degrees or so to avoid taking a zenith. Perhaps this is affecting the results?

Last edited by Dave_Banaszek (2007-06-25 19:23:44)

Dave_Banaszek wrote:

I'm going to stand by my assertion that the panohead is set up correctly though - because I can get a good quality stitch from that image set (without handheld nadir) without using smartblend or retouching in photoshop, aswell as the hundreds of other panoramas I've managed to stitch successfully since acquiring my fisheye lens using the same head settings.  I also seem to have arrived at the same settings as several other panoguide forums users with the same 350d-sigma8-nodal ninja setup

Hi Dave,

One way to prove that two of the three settings are correct is to insure the upright part of the Panohead is straight up and down, and the camera is pointing straight down (ie use a hotshoe level).  Take a nadir shot.  Rotate the head 180 degrees.  Take another shot.  Bring both images into photoshop and put one of them as a layer on the other.  Set the transparancy for the top one to 50%, and post the result.

We allow the crop circle to be not centered. Of course, distortion is calculated relatively to this center, so if the crop is not centered, distortion is not centered. Perhaps, there's something to study in this area.

Dave_Banaszek wrote:

I'm having difficulties producing a useable example for you here, but my best effort is shown below.  If I set the focus to the usual value I use for outdoor panos (halfway between 1 and infinity), the nadir image is out of focus so that it's hard to see in the photo whether the centre of the panohead shifts when moved through 180 degrees. However, changing the focus so that the image is sharp results in the centrepoint of the panohead moving in the viewfinder as the focussing ring is turned, so this wouldn't result in an accurate setting for the horizontal rail. I've tried the same experiment with my 17-40mm and the position in the viewfinder stays constant as the focus is changed. Perhaps this is the problem - my lens seems to be off-centre. I've noticed when setting the yellow crop circle that the image circle of the lens is offset slightly to the right. While this doesn't appear to affect the sharpness, due to the extensive depth of field, it does seem to cause problems correcting the distortion in APP.

OK, before you blame the lens, lets try a couple of things....

Nodal Ninja seems to have two adjustments, call them "fore and aft" and "up and down".  Fore and aft is the adjustment controlled by the screw that fits into the tripod mount on your camera, up and down is the other one.  (left and right are not needed because your tripod mount is centered with the lens).  These directions (up/down, left/right) are with respect to the camera sensor.

insure that the arm is straight up and down.  you could get a "z" type of pattern where the camera is level but offset in the left/right direction (does ptgui tell you if this is the case?)

Some lenses change the position of the NPP with focus changes, so your fore/aft setting may be off for panos, but not for this test. (because this test does not care about the fore/aft setting).

Set your aperture very small -- f/16 or even f/32 or however small the lens goes -- whatever gives you the best image.  Try several shots at different focus settings.   This will tell you if some element inside the lens is off center, but it won't prove that the lens is good.

Last edited by hankkarl (2007-06-26 21:58:17)

Dave_Banaszek wrote:

However, changing the focus so that the image is sharp results in the centrepoint of the panohead moving in the viewfinder as the focussing ring is turned, so this wouldn't result in an accurate setting for the horizontal rail.

For the following (1) and (2) adjustments I changed the focus to get a crisp image because I doubt focus moves the point where the lens axis cross the camera sensor. I will try to test that...

hankkarl   wrote:

Nodal Ninja seems to have two adjustments, call them "fore and aft" and "up and down". (left and right are not needed because your tripod mount is centered with the lens).  These directions (up/down, left/right) are with respect to the camera sensor.

The forward/backward is the third and last adjustment among three necessary adjustments.

The above image is an attempt to show the first two ones which, in my opinion, should be done before the third  "fore and aft" one.

The first one (1) is quite evident.
The second one (2) is less evident but useful to insure the lens axis  and the rotation vertical axis are exactly at the same place when the "Align" white arrow in the figure points on the 90 or on the 0 mark.

That you have to use the same (2) knob to deal with the third "fore and aft" adjustment when using a NN3 is an inconvenient the Panosurus avoids...

Note: both spirit levels being centered helps but using spirit levels is not mandatory.

I use initial value I established using a messy room for FOV a, b, d, e. I optimize for yaw, pitch and roll first. Then I optimize FOV. When I optimize a, b, d and e the changes are very low in most occasions, I sometimes skip this last optimization step.

Whether this is a common method or not, I don't know but doubt this is widespread (how to use optimizer is hard to explain, and, besides Johnh tutorial readers, not a very well known technique...)

Here's a sum up of improvement done by this adding ( and the fact that APP used a 2nd type of distortion, thing that no panotool tool can correct ).

These panoramas come from 8 to 10 picture shoot with a sigma 8mm / on canon 20D.

       |  std corr. | d/e/2nd type | clean CP>5.0 |    k1    |    k2    |  offx |  offy
       |            |              | + reoptimize |    k1    |    k2    |       |
----------------------------------------------------------------------------------------
pano 1 |       3.45 |         2.74 |        2.63  |  6.10^-3 | -4.10^-2 |  12.4 |    0
pano 2 |       3.37 |         2.61 |        2.58  |  9.10^-3 | -4.10^-2 |  13.1 |  3.9
pano 3 |       3.50 |         3.30 |        2.73  | -1.10^-2 | -4.10^-2 |   6.5 | -2.3
pano 4 |       3.49 |         2.74 |        2.60  |  8.10^-3 | -5.10^-2 |  14.3 | -7.74
pano 5 |       3.45 |         2.83 |        2.69  | -3.10^-2 | -3.10^-2 |   6.7 | -7.5

It always improves the result. That's a good thing
Notice : even if it's the same camera/lens, model of the lens varies a lot ... That's not good.

AlexandreJ wrote:

To Dave_Banaszek : Can I use this sample set to create a documentation in the wiki part about fisheye advanced lens model ?

Yes, certainly.