In a normal 3D renderer, a scene’s objects are sorted from back to front to ensure that occlusion and transparency work in the expected fashion.

We can deliberately mix this up by sorting scenes the right way (back to front) for the objects in the world, but the “wrong” way (front to back) for depth and motion.

Occluded objects are thus made visible in depth and motion layers, creating uncanny superimpositions as background objects “paint” ones in the foreground. This is an unreal but perhaps psychologically true way of rendering depth.