This talk will overview some recent research works on digital geometric signal processing, dynamic visibility culling, image-based modeling & rendering conducted at the State Key Laboratory of CAD & CG, Zhejiang University. First, a unified framework of geometric signal processing will be presented. A global parameterization of a mesh is achieved by projecting it onto a sphere. The geometric signals associated with each vertex of the mesh such as surface normal, color, reflectivity, texture coordinates etc. then turn spherical signals and can be converted into a spectral form by adopting the spherical harmonic transform, hence spectrally analyzed and synthesized to meet the requirements from various applications. Secondly, an efficient visibility culling approach for walkthroughs of complex virtual environments will be introduced. This approach is based on a compact representation, called Global Occlusion Map (GOM), for storing the global visibility information of the virtual environment with respect to each view cell. Finally, some recent research results on image-based visible hull reconstruction, garment texture change and synthesis of Chinese landscape will be presented.