Pose covariance

[Alessandro Dal Grande]

If I am optimizing the pose with Ceres, how can I get the final covariance?

Thanks

[Alessandro Dal Grande]

PS: ideally I am wanting to get a 6x6 covariance matrix of the current pose like in PTAM, if it's more helpful.

Thanks

[Sameer Agarwal]

By using the covariance class included with Ceres.

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/b94c18fd-c045-4029-8ecd-75806e643746%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Alessandro Dal Grande]

Hey Sameer,

I went and tried to implement like in the example, but I am left with a question: if I am optimizing one camera and the points are fixed, what should I use as covariance blocks?

Thanks a lot

[Alessandro Dal Grande]

Seeing what these guys did, they make a pair with the same camera?

https://github.com/arpg/Calibu/blob/master/include/calibu/calib/Calibrator.h#L287

[Sameer Agarwal]

you want the diagonal covariance block corresponding to your camera. so yes you ask for the covariance block corresponding to (camera, camera). does that make sense?

Sameer

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/c936c4cd-9337-4df3-ac62-5c051e53ed15%40googlegroups.com.

[Alessandro Dal Grande]

Yes, will test that.

Thanks

[Alessandro Dal Grande]

Hey Sameer,

I didn't get to the covariance yet, the coding is aborting on this:

CHECK(covarianceCalculator.Compute(covarianceBlocks, &problem));

My full code for this part is:

 ceres::Covariance::Options covarianceOptions;
 ceres::Covariance covarianceCalculator(covarianceOptions);
 vector<pair<const double*, const double*> > covarianceBlocks;
 double packedCamera[9];
 ...
 covarianceBlocks.push_back(make_pair(packedCamera, packedCamera));
 CHECK(covarianceCalculator.Compute(covarianceBlocks, &problem));

It seems it cannot find an item of some collection and then the logger (the one for Android/iOS) aborts as it is a fatal error.

What does it expect I feed it in? My problem blocks are originally set like this:

problem.AddResidualBlock(costFunction, lossFunction, packedCamera, packedPoint);

Thanks

[Sameer Agarwal]

can you post the full error log?

--

You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/236432a9-0088-4cbd-8e04-957c0dda6830%40googlegroups.com.

[Alessandro Dal Grande]

The log is as following:

map_util.h:62 Check failed: it != collection.end() Map key not found: 0x11e94e0

* thread #4: tid = 0x4875c, 0x39ea4dfc libsystem_kernel.dylib`__pthread_kill + 8, queue = '', stop reason = signal SIGABRT

    frame #0: 0x39ea4dfc libsystem_kernel.dylib`__pthread_kill + 8

    frame #1: 0x39f24d12 libsystem_pthread.dylib`pthread_kill + 62

    frame #2: 0x39e44938 libsystem_c.dylib`abort + 76

    frame #3: 0x004d0d40 MessageLogger::~MessageLogger() + 432

    frame #4: 0x0049a9c0 std::__1::map<double const*, int, std::__1::less<double const*>, std::__1::allocator<std::__1::pair<double const* const, int> > >::value_type::second_type const& ceres::FindOrDie<std::__1::map<double const*, int, std::__1::less<double const*>, std::__1::allocator<std::__1::pair<double const* const, int> > > >(std::__1::map<double const*, int, std::__1::less<double const*>, std::__1::allocator<std::__1::pair<double const* const, int> > > const&, std::__1::map<double const*, int, std::__1::less<double const*>, std::__1::allocator<std::__1::pair<double const* const, int> > >::value_type::first_type const&) + 240

    frame #5: 0x0049864a ceres::internal::CovarianceImpl::ComputeCovarianceSparsity(std::__1::vector<std::__1::pair<double const*, double const*>, std::__1::allocator<std::__1::pair<double const*, double const*> > > const&, ceres::internal::ProblemImpl*) + 794

    frame #6: 0x0049830c ceres::internal::CovarianceImpl::Compute(std::__1::vector<std::__1::pair<double const*, double const*>, std::__1::allocator<std::__1::pair<double const*, double const*> > > const&, ceres::internal::ProblemImpl*) + 68

  * frame #7: 0x00123cce +[BundleAdjustment optimizePose:withCorrespondencies:](self=0x005a7870, _cmd=0x00587b1c, keyframe=0x167b40a0, correspondencies=<unavailable>) + 5262 at BundleAdjustment.mm:214

Thanks

[Sameer Agarwal]

did you by chance mark the camera block as constant?

--

You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/e6827956-b50f-47b1-8977-82b5e4b21446%40googlegroups.com.

[Alessandro Dal Grande]

Part of it (the intrisics):

vector<int> constantIntrinsics;
 
  // set focal, px, py constant
  constantIntrinsics.push_back(6);
  constantIntrinsics.push_back(7);
  constantIntrinsics.push_back(8);
 
  ceres::SubsetParameterization *constantIntrinsicsParametrization = new ceres::SubsetParameterization(9, constantIntrinsics);

[Alessandro Dal Grande]

The points are also marked constant.

[Sameer Agarwal]

hmm that should not be a problem. can you share the full code fragment?

To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/4b3012c1-2f0f-452d-ba49-c889c11299bd%40googlegroups.com.

[Greg Anders]

I know this is an old thread, but did you ever figure out the solution? I am having the exact same problem (the same error of it != collection.end() Map key not found)

[Alessandro Dal Grande]

I think I solved it, but being years ago I unfortunately don't remember how. My suggestion is try to break down and simplify the problem as much as you can, and then add back logic step by step.