Ceres Autodiff for Kalman filter tuning

[Goksan Isil (Alumni)]

Hi!
I'm a relatively new user of Ceres and I'm currently trying to auto-tune the process noise matrix parameters (just the diagonal entries) of the standard Kalman filter via Ceres.
I have the ground truth bounding boxes, which are used for measurement update as well every 10th step. The states are the bounding box coordinates + velocity of the center of bbox.

I have the implementation as follows: link: https://github.com/goksanisil23/lazy_minimal_robotics/blob/main/Tracking/VisualTracking/KalmanErrorTerm.cpp

I would like to get some general feedback on the implementation, especially since I'm a bit confused about wrapping the numerical constants with the template type before assigning to my state vector during initialization. 

I initially wanted to add a residual block for each prediction step of the KF, but since the internal states of the filter at k, effects k+1, I created one residual block with N dimensions where N is the size of my ground truth dataset.

I'm open to suggestions since I'm quite sure this is not the right way to implement it. 
I'm currently getting 0 cost for example, and the optimization terminates immediately.

Thanks in advance!

[Goksan Isil (Alumni)]

Update: 0 cost problem was due to a typo so ignore that, I'm still open to feedback:)

[Sameer Agarwal]

I cannot access the link.

--
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/4ef97606-a754-47ff-95e1-7dfdf5b488a6n%40googlegroups.com.