[ptx] autopano-sift 1.3, libsift 1.1 released. (including RANSAC). RFC

[Sebastian Nowozin]

Hey panorama enthusiasts,


I am happy to announce a new version of my autopano-sift toolbox, version 1.3.
This version's changelog holds:

  + Add two dimensional affine transform model. This allows to represent the
    transformations necessary to planar rotations, scalings and translations
    between two image coordinate systems and is used as base for RANSAC.
  + RANSAC - RANdom SAmple Consensus. This allows accurate filtering of many
    matches between two images and filters geometrically implausible matches.
    RANSAC can be disabled in both the GUI (unmark RANSAC checkbox) and CLI
    ("--ransac off") for special lens geometries.
  + Add distance favoring metric for matching with RANSAC. This favors model
    fittings when the keypoints are spread out, leading to good coverage.
  + Variable downscaling of input image to requested dimension using
    hyperbolic filters, providing more controllable tuning to memory and speed
    requirements. Using the GDK library.
  + Add manual pages for all CLI programs, generate .txt, .ps, .pdf output
    from the pages.

  * Fix bug in the "autopano-complete.sh" script, thanks to Bruno Postle.


As usual, available from:
	http://cs.tu-berlin.de/~nowozin/autopano-sift/

I would be happy if people would test the new RANSAC option (works well for me)
and give feedback for situations where the RANSAC output differs vastly from
the non-RANSAC one, either in the good or bad.


Request For Comments
The underlying model that is used to evaluate the RANSAC fittings is shortly
described here:
	http://user.cs.tu-berlin.de/~nowozin/autopano-sift/matchmodel.pdf

I would love to hear feedback and ideas for improvements on the model from
anyone familar with geometric models or RANSAC or hugin. :-) (just send to the
list)


Thats all for today,
Sebastian

-- 
nowozin at cs.tu-berlin.de --- http://user.cs.tu-berlin.de/~nowozin/