Viktor Larsson

You are right, we could do default implementations with (`unproject` -> `project_with_jac` -> `pinv`) for camera models where we dont have a specialized the implementation. I agree that this is...

I think having it as `xd = K distort(xu)` is more natural. Otherwise you will either need to separate out the principal point from K or have the distortion compute...

We could potentially merge this branch + `camera_models` and [`absolute_pose_unknown_focal_clean`](https://github.com/PoseLib/PoseLib/tree/absolute_pose_unknown_focal_clean) into the `dev` branch first, to iron out any bugs before pushing to master. I think `camera_models` is fairly close...

I did a quick merge between `camera_models` and `absolute_pose_unknown_clean` onto `dev` now. I removed the `estimate_absolute_pose_focal`, and merged it into `estimate_absolute_pose` as this https://github.com/PoseLib/PoseLib/blob/dev/PoseLib/robust.cc#L50-L58

Yeah, let's merge it there first and sort out any issues.

Perhaps we should only keep the factorized version in `ProjectiveImagePair` and remove `Matrix3d`. Could expose it via something like `Eigen::Matrix3d ProjectiveImagePair::F() { return FF.F() }; ` This might require a...

> I thought about it now and maybe we can just create `FactorizedProjectiveImagePair` which is only used during optimization. Similar to how `FactorizedFundamentalMatrix` is used only within the optim module....

@kocurvik is this branch ready to merge now?

> I think it is almost ready. > > The only thing I am unsure about is the minus sign in the `jac_param` calculation in the `unproject_with_jac` which I think...

That would be great! Thanks for your contributions. There is a work-in-progress branch that is refactoring the camera models to be 2d-3d instead of 2d-2d, which includes the division model...