What's the "ground truth"

[Flix]

Hi,

the term is first used in l9 on page 70. I didn't attend the lecture and searching the internet get's me nothing I could relate to those nice and colorful pictures.

Thanks

[keksberg]

Hey,
in Computer Vision "ground truth" is something like the theoretical and optimal result of a problem and you can use it to benchmark Computer Vision methods.
This ground truth can either be calculated when using a synthetic benchmark or it is measured by hand or something else...
Usually you have different methods (here: pyramid LK method) to solve a specific problem (here: get the optical flow) and to evaluate this method, you compare it to this ground truth.
Another example: You want to reconstruct the geometry of an object from photos of it. To evaluate your algorithm, you reconstruct the geometry from the photos and compare the 3D model with an accurate 3D laser scan of the object. In this case, the model from the laser scan is your ground truth... how it actually should be.

[Flix]

Thanks a lot for your explanation, I assumed in this context the term was somehow directly related to the optical flow.