Description: We aim to individually evaluate each building reconstructed in 3D using Very High Resolution
(VHR) optical images. Detected errors will be ranked, and a human operator will then validate
or invalidate such errors. It will help us (1) robustifying the learning process of acceptable 3D
reconstructions, given a level of details, and (2) upscaling the process with a direct labeling of
new areas. Various 3D modeling techniques will be compared in order to assess which one(s)
is(are) the most suited for which urban environment.