As the basis for representation, Gabors work well for 2D (be it biological visual cortex or converged CNN kernels - the math on Gabors efficiency as such a basis explains that commonality of biology and CNN) and with beginning of our venturing into 3D - primarily with the recently starting boom in robotics - one can see that spherical harmonics - probably localized like Gabors - would take similar role of a such an efficient basis there.
alt Hacker News
As the basis for representation, Gabors work well for 2D (be it biological visual cortex or converged CNN kernels - the math on Gabors efficiency as such a basis explains that commonality of biology and CNN) and with beginning of our venturing into 3D - primarily with the recently starting boom in robotics - one can see that spherical harmonics - probably localized like Gabors - would take similar role of a such an efficient basis there.