dellaert
|
f3ee25f1a8
|
TriangulationFactor, first version
|
2014-03-02 14:56:50 -05:00 |
|
dellaert
|
dfee108e53
|
Some more refactoring
|
2014-03-02 13:49:42 -05:00 |
|
dellaert
|
bf779af3d1
|
added twoPoses test
|
2014-03-02 13:44:18 -05:00 |
|
dellaert
|
7b93cd207c
|
fixed header bloat
|
2014-03-02 13:34:43 -05:00 |
|
dellaert
|
68401cf216
|
removed Cheirality testing in that GTSAM mode
|
2014-03-02 13:16:36 -05:00 |
|
dellaert
|
ba6f857663
|
Removed some copy/paste
|
2014-03-02 13:02:21 -05:00 |
|
dellaert
|
744d9f7c1c
|
Cherry-picked Richard commit: Removed use of boost::range 'join' - replaced with a special flag to add one dimension in VerticalBlockMatrix and SymmetricBlockMatrix
|
2014-03-02 00:07:12 -05:00 |
|
dellaert
|
15a69fa1ca
|
Maybe this does not make a difference, and it's a bit ugly, but I was thinking this change would allow the methods in question to be more readily inlined by the compiler.
|
2014-03-01 19:26:58 -05:00 |
|
dellaert
|
98ae00f1ef
|
Avoid malloc in common case?
|
2014-03-01 19:24:51 -05:00 |
|
dellaert
|
fd376a5247
|
Avoid millions of mallocs when using Unit noise model
|
2014-03-01 18:10:21 -05:00 |
|
dellaert
|
e4edaf359c
|
Avoiding mallocs and working with fixed blocks where possible, makes a pretty noticeable difference in the inner linearize loop
|
2014-03-01 18:04:43 -05:00 |
hchiu
|
6b215ea8d9
|
Use reuse_diagonal_ with step policy to save computation time (same as Ceres).
|
2014-02-28 20:45:09 -05:00 |
|
hchiu
|
037ed7b931
|
Speed Optimization: Move sqrt computation to hessianDiagonal storation as Luca suggested. Got same values in unit tests (TestNonlinearOptimizer).
|
2014-02-28 18:09:10 -05:00 |
Luca
|
e127f07336
|
fixed iteration counter
|
2014-02-26 14:36:51 -05:00 |
|
Luca
|
a90c5797e0
|
lower bound on lambda in the right place
|
2014-02-26 12:56:41 -05:00 |
|
dellaert
|
c3f0bf4949
|
Fixed localCoordinates bug in all three Calibration objects.
|
2014-02-26 10:30:34 -05:00 |
|
Luca
|
49a1b7abca
|
partially fixed unit test on LM
|
2014-02-25 21:09:05 -05:00 |
|
Luca
|
6217a0b6c4
|
fixed LM policy, using CERES
|
2014-02-25 21:02:30 -05:00 |
|
Luca
|
6962072301
|
moved Jacobians on SO3 to Rot3
|
2014-02-24 14:00:14 -05:00 |
|
Luca
|
87687ff887
|
added simplest BAL file
|
2014-02-24 13:07:53 -05:00 |
|
Luca
|
c828bf567e
|
removed cout
|
2014-02-21 20:58:01 -05:00 |
|
Luca
|
98e32a1e6d
|
reorganized verbosity and included the new lambda policy from CERES
|
2014-02-20 17:17:12 -05:00 |
|
Luca
|
06c5186fa2
|
added function to create initial values from db
|
2014-02-20 15:37:17 -05:00 |
|
Luca
|
7e8bb1ffb0
|
rearranging comments
|
2014-02-20 15:36:11 -05:00 |
|
Luca
|
a0bc4b0a5a
|
added 1 more unit test: delta is perpendicular to gradient of original system
|
2014-02-20 12:35:07 -05:00 |
|
Luca
|
dec6430d9b
|
1 more test: a small step in delta direction does not decrease error
|
2014-02-20 12:22:12 -05:00 |
|
Frank Dellaert
|
541dbd5199
|
Plot thickens: delta computed from damped system points downhill but non-linear error does not decrease
|
2014-02-20 09:02:24 -05:00 |
|
Frank Dellaert
|
33ae83fcb2
|
Much more extensive testing of diagonal policy with Luca over Skype - gets stuck in a non-minimum !!!
|
2014-02-20 00:28:15 -05:00 |
|
Frank Dellaert
|
666072b169
|
Created new method buildDampedSystem with Luca
|
2014-02-20 00:27:33 -05:00 |
|
Frank Dellaert
|
53134425d5
|
Fixed verbosity check
|
2014-02-20 00:26:48 -05:00 |
|
Luca
|
9adc59af18
|
added verbosity for debugging termination
|
2014-02-18 17:33:00 -05:00 |
|
Luca
|
92834a8ed7
|
modified verbosity levels
|
2014-02-18 14:39:12 -05:00 |
|
Luca
|
fbfa287d4b
|
added parameters in "print" for LM
|
2014-02-17 17:43:45 -05:00 |
|
hchiu
|
13254bcc3a
|
Add related code for reuse_diagonal, as Ceres did. This is for step strategry (efficiency comparison) in smartFactors_Ceres.
|
2014-02-17 10:13:14 -05:00 |
|
Luca
|
c05eaa0d22
|
fixed unit test
|
2014-02-16 17:11:58 -05:00 |
|
Luca
|
4aec076568
|
do not cross upper bound for lambda
|
2014-02-16 16:33:57 -05:00 |
|
hchiu
|
a1ff716daf
|
Add min_diagonal and max_diagonal as Ceres.
|
2014-02-16 16:32:18 -05:00 |
|
Luca
|
9279d4402f
|
identifying convergence issue in diagonal hessian LM
|
2014-02-16 16:19:20 -05:00 |
|
Luca
|
8b6bb80cec
|
small fix
|
2014-02-16 16:05:01 -05:00 |
|
hchiu
|
f804fd9bd1
|
Traced Ceres and took sqrt of diagonal term as they did.
|
2014-02-16 15:50:26 -05:00 |
|
dellaert
|
e88b214b40
|
Small re-factor, no more finished()
|
2014-02-16 12:17:00 -05:00 |
|
dellaert
|
f8024bd6dd
|
Conversion from Hessian can have NULL node model
|
2014-02-16 12:15:52 -05:00 |
|
dellaert
|
17e2221395
|
New test showing diagonal damping currently broken.
|
2014-02-16 11:58:13 -05:00 |
|
dellaert
|
c31d48dba1
|
New flag to govern diagonal damping, off by default
|
2014-02-16 11:57:51 -05:00 |
|
dellaert
|
19a06ca68f
|
Added unit tests for NULL noise model and fixed bug in hessianDiagonalXX for that case
|
2014-02-16 00:21:15 -05:00 |
|
hchiu
|
b0618b8894
|
Replace the identity matrix with diagonal of Hessian.
|
2014-02-15 20:46:28 -05:00 |
|
Luca
|
e789553261
|
bug fix: hessianDiagonal seg-faulted with NULL factors
|
2014-02-15 16:48:15 -05:00 |
|
Luca
|
71e4015304
|
bug fix: before hessianDiagonal seg-faulted with no noise model
|
2014-02-15 16:44:20 -05:00 |
|
dellaert
|
ba6f8b7cca
|
Merge branch 'origin/feature/LM/SymmetricBlockMatrix'
|
2014-02-15 12:20:06 -05:00 |
|
dellaert
|
95c13d64dc
|
Minimal unit test, forgotten by @richardroberts :-)
|
2014-02-15 12:10:41 -05:00 |