How to compare results with the ground truth?

[simonfuhrmann]

How to compare results with the ground truth?

[simonfuhrmann]

This is unrelated to MVE and does not belong here. However, the proper way to do this is to use the camera information provided with Strecha's data sets and directly inject them in the MVE views. Since you need SfM points for MVS, this will not work "easily" as you have to generate SfM points yourself. Another way is to align the meshes using ICP.

[Meekohi]

@mojovski it is also reasonable to just run iterative closest points (ICP) to align the computed model to the ground truth, and then measure deviation based on that result. That is probably the quickest solution (MeshLab will do this for you, for example).

ok icp is an option. Thank you!

by the way: what would I have to do in order to inject known (or approximately known) camera poses into the sfm module? I usually have precomputed poses from a SLAM system. Is this somehow supported?

[simonfuhrmann]

You can edit the views (views/*.mve) by hand, or use the API to set the camera parameters.
https://github.com/simonfuhrmann/mve/blob/master/libs/mve/view.h#L120