How to get the Derivatives (Jacobians)

[YangSiri]

Hi, @jedeschaud

I think you've done a great job in the field of SLAM. The results of CT-ICP is significant. The coding is good too. Would you please tell me how to get the derivatives (jacobians) of CTPointToPlaneFunctor with respect to the corresponding two poses? I see you didn't use any form of cost function defined in ceres such as AutoDiffCostFunction. So i wonder how the optimization works by gradients?

https://github.com/jedeschaud/ct_icp/blob/1ba7ce704e9994d39076089ea3fc0dc4d856fe84/src/ct_icp/cost_functions.h#L71

[pdell-kitware]

Hi @YangSiri,

So we are indeed actually using the AutoDiffCostFunction,

However in the file you refered to, we only define the Cost Functors,

The cost functions are defined here:

https://github.com/jedeschaud/ct_icp/blob/1ba7ce704e9994d39076089ea3fc0dc4d856fe84/src/ct_icp/ct_icp.cpp#L229

Does this answer your question ?

Best regards,

[YangSiri]

Thanks, @pdell-kitware

I see your point. So i wonder whether you analytically infer the derivatives which i'm working on recently.

Very glad to receive your reply~

[chengwei0427]

Hi, @YangSiri ,have you analytically infer the derivatives? I'm working on it recently, and cannot get a good result.