Error with Inliers_RMSE

### Checklist

- [X] I have searched for [similar issues](https://github.com/isl-org/Open3D/issues).
- [X] For Python issues, I have tested with the [latest development wheel](http://www.open3d.org/docs/latest/getting_started.html#development-version-pip).
- [X] I have checked the [release documentation](http://www.open3d.org/docs/release/) and the [latest documentation](http://www.open3d.org/docs/latest/) (for `master` branch).

### My Question

I encountered an issue regarding the evaluation criteria of interactive visualization for ICP registration using either point-to-point or point-to-plane registration. When comparing the target and reference point cloud (PCDs), there seems to be a misunderstanding in the evaluation results.

Sometimes, while comparing meshes (identical meshes generated by different algorithms) with ground truth data, the mesh with the higher fitness score, also has the higher IRMSE. For example,
mesh1_VS_groundtruth: f-score=95.908% IRMSE=1.811e-02
mesh2_VS_groundtruth: f-score=95.030% IRMSE=1.782e-02

Is this expected to happen in some cases?
Based on theory:

F-score =Nc/Np ,
IRMSE =1/Nc*(∑[dp−q]) ,
Nc is the number of correspondences,
Np is the number of points in the target point cloud
and dp−q is the mean squared distance between the correspondences.
Fitness describes the overlapping area between the two point clouds. Inlier RMSE is the average of the mean
square point to point distances of the correspondences(Inliers). A good registration results in a high fitness value (in the range [0,1]) and a low inlier RMSE value.