n0odo wrote:

I tested the software on my widest lens the Canon 15mm (full frame fisheye) and it worked just great, stitched something everything else failed at. I figured that if you could stitch this you could stitch anything! Bought the software.
Went out and used my new Canon TS-E 17mm L lens. Easy shot, daylight, very level, just 3 images with a 50% overlap. It cannot stitch it to save it's life, not even with help, why??? Even Photomerge does better! Here is the pano from Photomerge:
Any ideas?
Thanks, Jeff...

Welcome to the forum...

Some ideas:

Too much overlap? - more than 30% is seldom beneficial and may cause problems.

A lot of the upper 2/3rds of the images is plain blue sky - APG will probably fail to make links in those areas.

Repetitive elements in the architecture of the church can confuse the SIFT-based control points detection system and this can lead to incorrect links being made - have you explored the links and control points in the Panoirama Editor's Control Points Editor? This tutorial may be helpful even if it relates to the previous version of the software:
http://www.autopano.net/wiki-en/action/ … _panoramas

If you'd like to make the images available for download from the Web - preferably as a ZIP file - I'd be happy to take a look and would then be able to offer more precise and better informed comments.

Last edited by mediavets (2011-02-28 12:45:47)

Thanks both,
I did try and adjust the points and failed! (but if Photomerge could do it on Auto?).
I did lift the front of the lens slightly for the top of the spire, camera was level on all planes.
The 3 images are:
http://www.elmstudio.com/images/one.jpg
http://www.elmstudio.com/images/two.jpg
http://www.elmstudio.com/images/three.jpg
Please let me know when you have them and I will take them down?
Sorry no zip on this work machine.
Regards, Jeff

n0odo wrote:

I did try and adjust the points and failed! (but if Photomerge could do it on Auto?).
I did lift the front of the lens slightly for the top of the spire, camera was level on all planes.

Hey Jeff!

OF COURSE you can´t use the T/S capabilities of your lens when shooting stitched! The stitcher assumes a 17mm lens. Not a shifted or tilted 17mm lens . .

Shifting and tilting means to shift or tilt the axis of the lens.

best, Klaus

Last edited by klausesser (2011-02-28 14:17:09)

Edit the exif manually to 1000mm or more before detection & it stitches perfectly by adding manual control points in APG 2.5

By not enabling lens distortion correction I got a good RMS. But that the resulting image (first screen capture) looks decent is only the result of the blender ghosts removal. As a proof I included the central part (second screen capture) where anti-ghosts was disabled...

I have a Really Right Stuff, full nodal point calibrated setup. All I did was lift the front a little, shift has never caused problems for the "free Canon Photostitch" or the Photoshop Photomerge or even Stitcher Pro. These have all been used on my images in the past with the Canon 24mm TS-E mk1, fully shifted up or down to keep building stright.
I was rotating around the nodal point, I did do a full rotation but gave up stitching when I could not even stitch 3 images!
To answer a question I captured +2 stops, -2 stops and correct exposure (motor drive and autobracket), move and shoot again, do a 360.
I can't figure out why your software is having so much problem with this?
You were able to do a difficult 360 I did with a 15mm that no one else could touch.
My images came out like Georges second image.
Jeff

Yes, I did calibrate for a shift up (it did make no differance) it was not much, just remove some grass and add some sky about 20% of the total movement. (remember the camera was horizontal). This problem has nothing to do with the shift, it would be the same for this lens in center.
It has to be this 17mm, look at this 360 from the 15mm, same setup for capture:
Cheers, Jeff.

digipano wrote:

here is the final image

well - looks a bit strange, does it . .

best, Klaus

klausesser wrote:

When you shoot for stitches you anyway don´t need TS at all - it absolutely makes no sense.

Like most stitchers (PTgui, hugin and more) Autopano places the source images on a sphere at positions which must exactly correspond to the sensor orientation when these source images where recorded. Shifting the lens upward is equivalent to move the sensor position downward, this results in a wrong placement of the source images on said sphere.

If you want to shift the lens you should not use any of these stitchers.

When one want to use these stitchers and avoid recording a large part of the ground he must use a 2 axis panohead and point the lens upward. This the way these stitchers are designed to work.

As a test of my above theory, I added a large (green) area in the bottom part of n0odo's source images so that the modified image center roughly corresponds to the lens axis and stitched them with Autopano lens distortion correction selected: gobal RMS is much better, easy to obtain and above all the overlapping areas of sources images really correspond.

QED!

Last edited by GURL (2011-02-28 18:44:34)

Wikipedia:
Article:       http://en.wikipedia.org/wiki/Tilt-and-shift_lens
Discussion: http://en.wikipedia.org/wiki/Talk:Persp … ntrol_lens
... both are interesting!

When shift amount is as large as 10 mm on a 24mm x 36mm sensor this could be used to make single row instead of double row panos, provided the shift amount is known with sufficient accuracy (?) and the stitcher includes "crop circle like" tool (?) to take the shift into account...