Abstract: This paper presents a visualization framework for cities in the form of aerial LiDAR (Light Detection and Ranging) point clouds. To provide interactive rendering for large data sets, the framework combines level-of-detail (LOD) technique with hierarchical hybrid representations of both point and polygon of the scene. The supporting structure for LOD is a multi-resolution quadtree (MRQ) hierarchy that is built purely out of input points. Each MRQ node stores separately a continuous data set for ground and building points that are sampled from continuous surfaces, and a discrete data set for independent tree points. The continuous data is first augmented with vertical quadrilateral building walls that are missing in original points owing to the 2.5D nature of aerial LiDAR. The continuous data is then spatially partitioned into same size subsets, based on which hybrid point-polygon structures are hierarchically constructed. Specifically, a polygon conversion operation replaces points of a subset forming a planar surface to a quadrilateral covering the same space, and a polygon simplification operation decimates wall quadrilaterals of a subset sharing the same plane to a single compact quadrilateral. Interactive hybrid visualization is retained by adapting a hardware-accelerated point based rendering with deferred shading. We perform experiments on several aerial LiDAR cities. Compared to visually-complete rendering [10], the presented framework is able to deliver comparable visual quality with less than 8% increase in pre-processing time and 2-5 times higher rendering frame-rates.