[ptx] Autopano 1.03 Beta 2

[Pablo d'Angelo]

Hi Alexandre,

> Hi everybody,
> 
> After a week in holiday, I'm back on autopano and found many bugs inside
> (in the sift detector and in the ransac stage).
> 
> Now, I think I'm quite over all of them and the improvement beetween the
> 1.03beta
> and this release are really good. On my testing cases, I get no false points
> any
> more ( lens ranging from 8 mm to 300 mm, nature, architecture, etc).
> I'm now truly invariant in scalling and rotating. Just peek a picture, scale
> it,
> Rotate it and apply autopano against the original and you will see.
> 
> Where ?
>   http://www.le-geo.com/kolor/autopano/autopano_v103beta2.zip
> 
> Any feedback are as always welcome.

I think the following points are nice for integration into the frontends:

1. option to output only one panorama, as the frontends
   (hugin/ptgui/ptassembler) do not support the recognizing panorama
   paradigma, and will ignore the control points written into the second
   image.

2. for case 1, it would be nice to be able to specify an output file.

3. The control points in the project files are not subpixel accurate,
   but rounded to integers.

4. Usually the images of one project share the same lens settings. It would
   be nice if the project files would reflect that, example:

Instead of:

--------------------------
#-imgfile 1600 1200 "sta_3135.jpg"
o f0 y+0.000000 r+0.000000 p+0.000000 u20 d0.000000 e0.000000 v70.000000 a0.000000 b0.000000 c0.000000
#-imgfile 1600 1200 "stb_3136.jpg"
o f0 y+0.000000 r+0.000000 p+0.000000 u20 d0.000000 e0.000000 v70.000000 a0.000000 b0.000000 c0.000000
--------------------------

use

--------------------------
#-imgfile 1600 1200 "sta_3135.jpg"
o f0 y+0.000000 r+0.000000 p+0.000000 u20 d0.000000 e0.000000 v70.000000 a0.000000 b0.000000 c0.000000
#-imgfile 1600 1200 "sta_3136.jpg"
o f0 y+0.000000 r+0.000000 p+0.000000 u20 d=0 e=0 v=0 a=0 b=0 c=0
--------------------------

if the images are of the same size. Otherwise hugin will not recognize that
the variables are linked, and treat every images distortion/fov value
separately.


I can compile a linux version as well, if you want to.

How does the PCA-SIFT compare to the normal SIFT? Is there a big difference?

ciao
  Pablo