This is a quite interesting area because it sits between the more traditional (pure 3d mesh based) models [1] (which are wonderful for 3d contours and precision

This is a quite interesting area because it sits between the more traditional (pure 3d mesh based) models [1] (which are wonderful for 3d contours and precision, but typically don't handle textures, lighting etc) and modern GANs which "simply" deform images (i.e. deep fakes, style transfer). The challenge is that the deformation is not straight-forward at all! That's why it truly helps (as this paper shows) to use some 3d knowledge in aiding the deformation. But yes, I'd love to see the code for this. Shame it's not available. [1] i.e. something like STAR from MPI: https://star.is.tue.mpg.de/ https://star.is.tue.mpg.de/