Adventures with old scanned photos continue. This time quite interesting case.
Three source photos, images 1 and 2 nearly identical, image 3 portrait mode.
Three pano compilers used:
1. ptgui - result not so good
2. photoshop - result nearly satisfactory
3. autopano giga -
result nearly satisfactory
This time I added vertical cp's in ptgui. And used individual lens and shift parameters all images as Joost suggested. (btw, this is a very good hint, I have used it in other panos quite succesfully.)
The main reason for ptgui's bad result is not finding control points. It found 1 cp between image1 and image3, 4 cps between image2 and image3. Autopano found 0 cps between image2 and image3, but 44 cps between image1 and image3!
The fact that the buildings in image 3 are out of focus may have something to do this.
Autopano giga is discontinued and out of business. However it has one feature that I find useful. It can find possible panoramas within a folder. I found this pano from a folder that included more than 400 images.
I have always wondered how the control points are calculated. This case did not help it at all ;-)