ca2aa0cfd4
Refactoring to get grouped factors to compile again
2015-06-16 20:33:33 -07:00
82f6ca64a5
Merge remote-tracking branch 'origin/develop' into fix/zeroRowInQR
2015-06-15 01:09:19 -07:00
b68f763fe7
Merged in feature/LinearSolverCleanup (pull request #148 )
...
This has
- Separate Scatter class
- change from updateATA to updateHessian as a virtual method
- Fixed-size BinaryJacobianFactor that overloads updateHessian
- SFM Timing Script that takes BAL files
2015-06-15 01:08:02 -07:00
33e412f2ee
Code review
2015-06-15 01:05:48 -07:00
f523db2d98
IMPORTANT CHANGE/BUGFIX: QR elimination now (I think properly) leaves a row of zeros if the RHS after QR contains a non-zero. A corresponding change in the error calculation makes that Jacobian and Hessian factors now agree on error. I think this was a bug, because it affected the error, but (I think) it only pertained to "empty" JacobianFactors which have no bearing on optimization/elimination.
2015-06-15 00:55:53 -07:00
83171b6096
New example
2015-06-14 23:07:28 -07:00
51433c1ec2
Merged in feature/cleanupBase (pull request #154 )
...
Some cleanup in base
2015-06-14 20:42:13 -07:00
75e072396c
Refactored and renamed some internals
2015-06-14 16:05:39 -07:00
15966a65f2
Small reformat
2015-06-14 16:05:18 -07:00
a18875b598
Changed headers to GTSAM-style (near to far)
2015-06-14 16:05:09 -07:00
5771664622
Starting to diagnose issue with lower-left entry of Hessian
2015-06-14 16:01:54 -07:00
7698c52ce9
Created BinaryJacobianFactor template
2015-06-14 15:50:15 -07:00
06902209b0
Fixed bug in hessian scalar computation
2015-06-14 15:19:55 -07:00
30104a114e
More tests with failing example
2015-06-14 15:03:44 -07:00
a94c2e7323
Renamed to BinaryJacobianFactor
2015-06-14 15:02:48 -07:00
9fcd498d6a
BORG formatting
2015-06-14 13:37:51 -07:00
8c22684bbb
Went back to base 1, and used constructors for blocks (cleaner)
2015-06-14 13:02:44 -07:00
df226fc436
No longer store my own matrices but get same performance using block
2015-06-14 12:54:18 -07:00
e045a5e1f7
Added more powerful tests on updateHessian
2015-06-14 12:53:20 -07:00
850501ed52
BORG Formatting
2015-06-14 11:16:54 -07:00
2c99f68ed7
Some formatting/cleanup before fixing bug
2015-06-14 10:56:22 -07:00
3e1f336ede
Merged 'develop'.
2015-06-14 11:41:03 -04:00
4909fef21a
Fixed issue
2015-06-13 20:20:33 -07:00
4e1a0edb0a
Reverted back to vector to avoid troubles w TBB
2015-06-13 20:19:44 -07:00
107ed37e0c
Merge branch 'justKeys' into feature/LinearSolverCleanup
2015-06-13 16:23:45 -07:00
d0775faeba
Save slots to bring cost down from O(n^3) to O(n^2) - again, in theory. In practice, it did seem to help for larger HessianFactors (as expected).
2015-06-13 12:26:10 -07:00
f6575323d6
Sidestep Scatter altogether and just use HessianFactor keys_. In theory, n^3 lookup cost, but in practice (as keys is contiguous memory) just as fast as map.
2015-06-13 12:06:13 -07:00
08f30966dd
Got rid of obsolete getSlots method
2015-06-13 11:03:12 -07:00
6f78e00812
Merge branch 'slotMethod' into feature/LinearSolverCleanup
2015-06-13 09:17:18 -07:00
c8cff296fb
Don't bother making an array
2015-06-13 09:01:13 -07:00
68d9905359
Merged in feature/stereocamera_backproject_addJacobian (pull request #153 )
...
Add optional jacobians to StereoCamera backprojection
2015-06-12 17:29:26 -04:00
abe6b9cec6
change 'world' to 'point'
2015-06-12 17:27:36 -04:00
3d18d70d69
change naming for local_point & world_point
2015-06-12 16:55:38 -04:00
8440e3c3b2
cool, a simplified D_local_z jacobian
2015-06-12 16:53:32 -04:00
2a2b885cdd
change local variable naming...
2015-06-12 16:07:36 -04:00
950cdec3b6
Merged 'develop'.
2015-06-12 15:34:05 -04:00
b4ec622525
Merged in fix/testRegularJacobianFactor (pull request #152 )
...
fix the RegularJacobianFactor overload issue, for mutiplyHessianAdd function.
2015-06-12 15:23:52 -04:00
4c9b1de675
Tests passed with real settings
2015-06-12 15:09:28 -04:00
9ac223ec7d
correct the chain rule in Jacobian
2015-06-12 15:08:58 -04:00
4c4c72adb4
oops, this should be derivative against pose in test
2015-06-12 13:56:47 -04:00
fb34c099ec
fix the RegularJacobianFactor overload issue, for mutiplyHessianAdd function.
2015-06-12 13:35:45 -04:00
6664699c4c
getSlots method
2015-06-12 09:33:55 -07:00
a5d74f77d7
add test case. correct constness for backproject2 definition
2015-06-12 10:47:02 -04:00
b712a65c21
Updated gttic strings as well
2015-06-12 00:28:15 -07:00
41a0146b05
changed updateATA -> updateHessian. Much clearer
2015-06-12 00:20:37 -07:00
c9910625c2
Fixed headers
2015-06-11 23:51:35 -07:00
bcc34d1c12
No more copy constructor
2015-06-11 23:49:28 -07:00
15d5542809
Renamed variable
2015-06-11 23:48:18 -07:00
692ddd8f36
Made node keys ordered to avoid copy constructor in eliminate
2015-06-11 23:38:57 -07:00
63d918ed7a
Now FastVector
2015-06-11 23:38:29 -07:00
Frank Dellaert
dellaert
zhaoyang