search

RainerKuemmerle / g2o

g2o: A General Framework for Graph Optimization
3.07k stars 1.1k forks source link

Question about SE3 Jacobian matrix computation in g2o #146

Open hlc111 opened 7 years ago

hlc111 commented 7 years ago

Hi Author,

Sorry to bother you! Those days i'm study G2O, it's a great optimizer project. For more deep understand G2O, i'm debugging the code now. But for how to get Jacobian i can't understand in Jacobian computation of EdgeSE3 linearizeOplus computeEdgeSE3Gradient

Could you kindly explain or provide some math formula reference help me understand it?

Currently i can understand use rotation matrix get Quaternion w, but don't know why use qw get dq_dR and use dq_dR and use those data get Ji and Jj.

Thanks!

lianbin commented 3 years ago

Hi Author,

Sorry to bother you! Those days i'm study G2O, it's a great optimizer project. For more deep understand G2O, i'm debugging the code now. But for how to get Jacobian i can't understand in Jacobian computation of EdgeSE3 linearizeOplus computeEdgeSE3Gradient

Could you kindly explain or provide some math formula reference help me understand it?

Currently i can understand use rotation matrix get Quaternion w, but don't know why use qw get dq_dR and use dq_dR and use those data get Ji and Jj.

Thanks!

Hi,Do you know how to derive Jacobi now ? Can you tell me?

RainerKuemmerle commented 3 years ago

You may have a look at chain rule and friends. All together this is a detail of implementation and you may obtain the values of the Jacobian by other means, i.e., the default numerical implementation or by AD (as we are currently working on in a branch, see https://github.com/RainerKuemmerle/g2o/blob/rainer/auto-diff/g2o/core/auto_differentiation.h )