Hiya,
we just spent all night tinkering our projection, trying to match the actual architecture as close as possible. Thanks to MadMapper, the task was rather easy, although we found a few really annoying UI bugs. Due to some modelling inaccuracies, I had to split the projection in 3 parts, each perspective corrected. Nevertheless, we ended up being roughly satisfied (but ya know the tune… I can’t get no …).
Then we took a few photos on location. And as usual with video mapping, the parts you design the quickest are the best…
Here’s the building relighted in 3d (Ambient Occlusion + Soft Shadows + Toon shader):

And the headshot, fully voice-activated (AO + SS + 8 morph targets depending on the tone of the voice):

And the MadMapper setup:

Tomorrow, I’ll try another Madmapper technique, using either grid warped (but still perspective corrected) or maybe camera pose estimation (with an openGL matrix output-to-file feature, then back in QC).