Image-based rendering takes advantage of the bounded display resolution to limit the rendering complexity for very large datasets. However, image-based rendering also suffers from several drawbacks that polygon-based rendering does not. These include the inability to change the illumination and material properties of objects, screen-based querying of object-specific properties in databases, and unrestricted viewer movement without visual artifacts such as visibility gaps. View-dependent rendering has emerged as another solution for hierarchical and interactive rendering of large polygonbased visualization datasets. In this paper we study the relative advantages and disadvantages of these approaches to learn how best to combine these competing techniques towards a hierarchical, robust, and hybrid rendering system for large data visualization.