1 files changed, 92 insertions, 0 deletions
diff --git a/extern/libmv/third_party/ssba/README.TXT b/extern/libmv/third_party/ssba/README.TXT
new file mode 100644
index 00000000000..734962b1df8
--- /dev/null
+++ b/extern/libmv/third_party/ssba/README.TXT
@@ -0,0 +1,92 @@
+Description
+
+This is an implementation of a sparse Levenberg-Marquardt optimization
+procedure and several bundle adjustment modules based on it. There are three
+versions of bundle adjustment:
+1) Pure metric adjustment. Camera poses have 6 dof and 3D points have 3 dof.
+2) Common, but adjustable intrinsic and distortion parameters. This is useful,
+   if the set of images are taken with the same camera under constant zoom
+   settings.
+3) Variable intrinsics and distortion parameters for each view. This addresses
+   the "community photo collection" setting, where each image is captured with
+   a different camera and/or with varying zoom setting.
+
+There are two demo applications in the Apps directory, bundle_common and
+bundle_varying, which correspond to item 2) and 3) above.
+
+The input data file for both applications is a text file with the following
+numerical values:
+
+First, the number of 3D points, views and 2D measurements:
+<M> <N> <K>
+Then, the values of the intrinsic matrix
+    [ fx  skew  cx ]
+K = [  0   fy   cy ]
+    [  0    0    1 ],
+and the distortion parameters according to the convention of the Bouget
+toolbox:
+
+  <fx> <skew> <cx> <fy> <cy> <k1> <k2> <p1> <p2>
+
+For the bundle_varying application this is given <N> times, one for each
+camera/view.
+Then the <M> 3D point positions are given:
+
+  <point-id> <X> <Y> <Z>
+
+Note: the point-ids need not to be exactly from 0 to M-1, any (unique) ids
+will do.
+The camera poses are given subsequently:
+
+  <view-id> <12 entries of the RT matrix>
+
+There is a lot of confusion how to specify the orientation of cameras. We use
+projection matrix notation, i.e. P = K [R|T], and a 3D point X in world
+coordinates is transformed into the camera coordinate system by XX=R*X+T.
+
+Finally, the <K> 2d image measurements (given in pixels) are provided:
+
+  <view-id> <point-id> <x> <y> 1
+
+See the example in the Dataset folder.
+
+
+Performance
+
+This software is able to perform successful loop closing for a video sequence
+containing 1745 views, 37920 3D points and 627228 image measurements in about
+16min on a 2.2 GHz Core 2. The footprint in memory was <700MB.
+
+
+Requirements
+
+Solving the augmented normal equation in the LM optimizer is done with LDL, a
+Cholsky like decomposition method for sparse matrices (see
+http://www.cise.ufl.edu/research/sparse/ldl). The appropriate column
+reordering is done with COLAMD (see
+http://www.cise.ufl.edu/research/sparse/colamd). Both packages are licensed
+under the GNU LGPL.
+
+This software was developed under Linux, but should compile equally well on
+other operating systems.
+
+-Christopher Zach (cmzach@cs.unc.edu)
+
+/*
+Copyright (c) 2008 University of North Carolina at Chapel Hill
+
+This file is part of SSBA (Simple Sparse Bundle Adjustment).
+
+SSBA is free software: you can redistribute it and/or modify it under the
+terms of the GNU Lesser General Public License as published by the Free
+Software Foundation, either version 3 of the License, or (at your option) any
+later version.
+
+SSBA is distributed in the hope that it will be useful, but WITHOUT ANY
+WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
+A PARTICULAR PURPOSE.  See the GNU Lesser General Public License for more
+details.
+
+You should have received a copy of the GNU Lesser General Public License along
+with SSBA. If not, see <http://www.gnu.org/licenses/>.
+*/