To be honest, it is a question for computer vision.
There is a normal vector map $N(x,y,3)$ represents the normal vector $(nx,ny,nz)$ on the point $P(x,y)$. If the resolution is high enough (2448x2048), can we get the approximate mean curvature by this normal vector map?
I found a proof about the relationship between normal vector and curvature, it says $$\operatorname{div} (N)=\kappa_1+\kappa_2,$$ where $\kappa_1, \kappa_2$ are the principle curvatures. See here. It is easy to get the part of $x$ and $y$, but not $z$ since there is no $z$ coordinate.