Those are two largely unrelated problems. Reconstructing geometry from multiple views is a separate (very well understood) problem from analysis to understand what the shapes mean.
Not to mention for Tesla having to abandon all of their training data.