This paper presents a new approach for efficient 3D rendering on mobile devices, where selective rendering can be achieved with the help of 3D scene annotation. By taking advantage of first person environments in most 3D applications, we are able to annotate the flooring details of the 3D space. This allows 3D environments to be interfaced using a higher level view of objects. With the higher level of scene understanding, it is possible to determine which 3D objects are not required for loading or rendering based on the viewer’s location and its surrounding constraints.