about the computation of jacobian

[amiltonwong]

Hi, @halismai ,

I'm a bit confused about the computation of jacobian ( line.77 to 87) . It is very different to general analytic jacobian. Could you elaborate a bit about the computation of jacobian in bitplanes or show me where could I find the details in your papers?

THX~

[halismai]

The code implements a standard warp using 3D coordinates, which is given as the product of the image gradient and the warp jacobian, J = grad I * grad w.

The form of the Jacobian in the document differs in two ways:

Firstly, the product is written out explicitly in the document. Whereas in the code, we pre-compute the warp Jacobian to save computational time when using multiple channels.

Secondly, the form of the Jacobain in the document is specific to the disparity space, which is slightly different than using 3D points with a rigid-body transform.

I'll attach a Matlab script with the computation of the Jacobian later tonight.

[amiltonwong]

@halismai , would you attach the script with the computation of the Jacobian?

THX!

[halismai]

Attached are scripts to compute the Jacobian for SE(3) (the bvpo https://github.com/halismai/bpvo library).

For SL(3), the Homography, you will have to change the generators slightly. I'm looking for the scripts for the SL(3) generators and I'll send them to you as soon as I find them!

sym.zip

[amiltonwong]

Thanks @halismai , I'll have a check for it.