[Bf-blender-cvs] SVN commit: /data/svn/bf-blender [56656] branches/soc-2011-tomato/extern/ libmv: Update libmv to version from own branch
Sergey Sharybin
sergey.vfx at gmail.com
Fri May 10 14:35:31 CEST 2013
Revision: 56656
http://projects.blender.org/scm/viewvc.php?view=rev&root=bf-blender&revision=56656
Author: nazgul
Date: 2013-05-10 12:35:31 +0000 (Fri, 10 May 2013)
Log Message:
-----------
Update libmv to version from own branch
Brings in the changes:
- Optimized keyframe selection, which is reached by
switching pseudoInverse from SVD to eigensolver.
- Cleans up comments and functions names.
Also cleaned code in libmv-capi a bit.
Modified Paths:
--------------
branches/soc-2011-tomato/extern/libmv/ChangeLog
branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/bundle.cc
branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/keyframe_selection.cc
branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/keyframe_selection.h
branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/reconstruction_scale.cc
branches/soc-2011-tomato/extern/libmv/libmv-capi.cpp
Modified: branches/soc-2011-tomato/extern/libmv/ChangeLog
===================================================================
--- branches/soc-2011-tomato/extern/libmv/ChangeLog 2013-05-10 12:27:18 UTC (rev 56655)
+++ branches/soc-2011-tomato/extern/libmv/ChangeLog 2013-05-10 12:35:31 UTC (rev 56656)
@@ -1,3 +1,110 @@
+commit 42da053c6410b4f3fb13798c7e9c5f4a861b6825
+Author: Sergey Sharybin <sergey.vfx at gmail.com>
+Date: Fri May 10 18:30:40 2013 +0600
+
+ Keyframe selection improvements
+
+ Added additional criteria, which ignores
+ keyframe pair if success intersection factor
+ is lower than current candidate pair factor.
+
+ This solves issue with keyframe pair at which
+ most of the tracks are intersecting behind the
+ camera is accepted (because variance in this
+ case is really small),
+
+ Also tweaked generalized inverse function,
+ which now doesn't scale epsilon by maximal
+ matrix element. This gave issues at really bad
+ candidates with unstable covariance. In this
+ case almost all eigen values getting zeroed
+ on inverse leading to small expected error.
+
+commit f3eb090f7240f86799099fe86ce9386eb2bd3007
+Author: Sergey Sharybin <sergey.vfx at gmail.com>
+Date: Fri May 10 17:59:40 2013 +0600
+
+ Keyframe selection code cleanup
+
+ - Updated comments in code.
+ - Removed currently unused functions.
+ Actually, they'd better be moved to some generic
+ logging module, but we don't have it now so was
+ lazy to create one now. Could happen later using
+ code from git history
+ - Renamed function to match better to what's going
+ on in it.
+
+commit b917b48bd877eedd17dec907cacf0b27a36e717d
+Author: Sergey Sharybin <sergey.vfx at gmail.com>
+Date: Fri May 10 17:44:49 2013 +0600
+
+ Optimization for reconstruction variance
+
+ Achieved by replacing SVD-based pseudo-inverse with
+ an eigen solver pseudo inverse.
+
+ New function works in the same way: it decomposes
+ matrix to V*D*V^-1, then inverts diagonal of D
+ and composes matrix back.
+
+ The same way is used to deal with gauges - last
+ 7 eigen values are getting zeroed.
+
+ In own tests gives approx 2x boost, without
+ visible affect on selected keyframe quality.
+
+commit 041b4b54fff66311347a307a5922c2516c76ee44
+Author: Sergey Sharybin <sergey.vfx at gmail.com>
+Date: Thu Mar 14 14:53:42 2013 +0600
+
+ Initial commit of reconstruction variance criteria
+ which is an addition for GRIC-based keyframe selection.
+
+ Uses paper Keyframe Selection for Camera Motion and Structure
+ Estimation from Multiple Views,
+ ftp://ftp.tnt.uni-hannover.de/pub/papers/2004/ECCV2004-TTHBAW.pdf
+ as a basis.
+
+ Currently implemented camera positions reconstructions,
+ bundle positions estimation and bundle adjustment step.
+
+ Covariance matrix is estimating using generalized inverse
+ with 7 (by the number of gauge freedoms) zeroed eigen values
+ of J^T * J.
+
+ Additional changes:
+ - Added utility function FundamentalToEssential to extract
+ E from F matrix, used by both final reconstruction pipeline
+ and reconstruction variance code.
+
+ - Refactored bundler a bit, so now it's possible to return
+ different evaluation data, such as number of cameras and
+ points being minimized and also jacobian.
+
+ Jacobian currently contains only camera and points columns,
+ no intrinsics there yet. It is also currently converting to
+ an Eigen dense matrix. A bit weak, but speed is nice for
+ tests.
+
+ Columns in jacobian are ordered in the following way:
+ first columns are cameras (3 cols for rotation and 3 cols
+ for translation), then goes 3D point columns.
+
+ - Switched F and H refining to normalized space. Apparently,
+ refining F in pixel space squeezes it a lot making it wrong.
+
+ - EuclideanIntersect will not add point to reconstruction if
+ it happened to be behind the camera.
+
+ - Cleaned style a bit.
+
+commit 94c4654f1145f86779bd0a7e8cda1fd2c751d27d
+Author: Sergey Sharybin <sergey.vfx at gmail.com>
+Date: Fri May 10 13:27:21 2013 +0600
+
+ Left extra debugging print in reconstruction scale by accident.
+
commit 3886b488575ec5e79debce7b9f2e9824f5297c0f
Author: Sergey Sharybin <sergey.vfx at gmail.com>
Date: Fri May 10 12:23:03 2013 +0600
@@ -559,91 +666,3 @@
EuclideanBundle:
- Fix RMSE logging.
-
-commit 1696342954614b54133780d74d6ee0fbcbe224f0
-Author: Sergey Sharybin <sergey.vfx at gmail.com>
-Date: Tue Feb 26 18:10:33 2013 +0600
-
- Upgrade ceres to latest upstream version
-
- This is needed because of some features of new Ceres
- for further development.
-
-commit 575336f794841ada90aacd783285014081b8318c
-Author: Sergey Sharybin <sergey.vfx at gmail.com>
-Date: Mon Jan 7 15:58:40 2013 +0600
-
- Fixed for keyframe selection
-
- - Calculate residuals for GRIC in pixel space rather than
- in normalized space.
-
- This seems to be how it's intended to be used.
-
- Algebraic H and F will still use normalized coordinates which
- are more stable, after this matrices are converted to pixel
- space and Ceres refinement happens in pixel space.
-
- - Standard deviation calculation was wrong in GRIC. It shouldn't
- be deviation of residuals, but as per Torr it should be deviation
- of measurement error, which is constant (in our case 0.1)
-
- Not sure if using squared cost function is correct for GRIC,
- but cost function is indeed squared and in most papers cost
- function is used for GRIC. After some further tests we could
- switch GRIC residuals to non-squared distance.
-
- - Bring back rho part of GRIC, in unit tests it doesn't make
- sense whether it's enabled or not, lets see how it'll behave
- in real-life footage.
-
- - Added one more unit test based on elevator scene and manual
- keyframe selection.
-
-commit 24117f3c3fc5531beb6497d79bb6f1780a998081
-Author: Sergey Sharybin <sergey.vfx at gmail.com>
-Date: Sun Jan 6 19:07:06 2013 +0600
-
- Added test for keyframe selection based on manual selection
-
- Additional changes:
-
- - Reduce minimal correspondence to match real-world manually
- tracked footage
-
- - Returned back squares to SymmetricEpipolarDistance and
- SymmetricGeometricDistance -- this is actually a cost
- functions, not distances and they shall be squared.
-
-commit 770eb0293b881c4c419c587a6cdb062c47ab6e44
-Author: Sergey Sharybin <sergey.vfx at gmail.com>
-Date: Fri Dec 21 00:43:30 2012 +0600
-
- Improvements for keyframe selection
-
- - Changed main keyframe selection cycle, so in cases there're no
- more next keyframes for current keyframe could be found in the
- image sequence, current keyframe would be moved forward and
- search continues.
-
- This helps in cases when there's poor motion in the beginning
- of the sequence, then markers becomes occluded. There could be
- good keyframes in the middle of the shot still.
-
- - Extended keyframe_selection_test with real world cases.
-
- - Moved correspondences constraint to the top, so no H/F estimation
- happens if there's bad correspondence. Makes algorithm go a bit
- faster.
-
- Strangely, but using non-squared distances makes neighbor frames
- test fail, using squared distances makes this tests pass.
-
- However, using non-squared distances seems to be working better
- in real tests i've been doing. So this requires more investigation/
-
-commit 7415c62fbda36c5bd1c291bc94d535a66da896d0
-Author: Sergey Sharybin <sergey.vfx at gmail.com>
-Date: Thu Dec 20 18:46:09 2012 +0600
-
- Cosmetic change to correspondences reports in keyframe selection
Modified: branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/bundle.cc
===================================================================
--- branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/bundle.cc 2013-05-10 12:27:18 UTC (rev 56655)
+++ branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/bundle.cc 2013-05-10 12:35:31 UTC (rev 56656)
@@ -93,19 +93,29 @@
T predicted_x, predicted_y;
- // Apply distortion to the normalized points to get (xd, yd).
- // TODO(keir): Do early bailouts for zero distortion; these are expensive
- // jet operations.
- ApplyRadialDistortionCameraIntrinsics(focal_length,
- focal_length,
- principal_point_x,
- principal_point_y,
- k1, k2, k3,
- p1, p2,
- xn, yn,
- &predicted_x,
- &predicted_y);
+ // EuclideanBundle uses empty intrinsics, which breaks undistortion code;
+ // so use an implied focal length of 1.0 if the focal length is exactly
+ // zero.
+ // TODO(keir): Figure out a better way to do this.
+ if (focal_length != T(0)) {
+ // Apply distortion to the normalized points to get (xd, yd).
+ // TODO(keir): Do early bailouts for zero distortion; these are expensive
+ // jet operations.
+ ApplyRadialDistortionCameraIntrinsics(focal_length,
+ focal_length,
+ principal_point_x,
+ principal_point_y,
+ k1, k2, k3,
+ p1, p2,
+ xn, yn,
+ &predicted_x,
+ &predicted_y);
+ } else {
+ predicted_x = xn;
+ predicted_y = yn;
+ }
+
// The error is the difference between the predicted and observed position.
residuals[0] = predicted_x - T(observed_x);
residuals[1] = predicted_y - T(observed_y);
Modified: branches/soc-2011-tomato/extern/libmv/libmv/simple_pipeline/keyframe_selection.cc
===================================================================
@@ Diff output truncated at 10240 characters. @@
More information about the Bf-blender-cvs
mailing list