Revealing carved parts in rock art is of primary importance and remains a major challenge for archaeological documentation. Computational geometry applied to 3D imaging provides a unique opportunity to document rock art. This study evaluates five algorithms and derivatives used to compute ambient occlusion and sky visibility on 3D models of Mongolian stelae, also known as deer stones. By contrast with the previous companion work, models are processed directly in 3D, without preliminary projection. Volumetric obscurance gives the best results for the identification of carved figures. The effects of model resolution and parameters specific to ambient occlusion are then evaluated. Keeping tridimensional information intact allows accurate measurement of distance, volume, and depth. Objects augmented by ambient occlusion can easily be manipulated in 3D viewers, for seamless and effortless access to the overall organization of the figures, at the scale of the entire object. Qualitatively speaking, the 2D projected outputs are equivalent to, or even better than, existing archaeological documentation. The proposed workflow should be easily applicable in many situations, particularly as the functions provided for the free R programming software are fully configurable.