Mesh compression method with on-the-fly decompression during rasterization and streaming support

In this article we propose a method of mesh compression and streaming, that can be used for real-time rendering applications. While most of other existing compression methods require decompression on CPU before rendering and streaming methods use only non-compressed models, our approach allows to reduce memory consumption by applying decompression on GPU during rendering and streaming of only needed geometry data at the same time. Proposed approach requires pre-processing step, on which coarse 3D model with quad faces is build and resampling is done. Afterwards, each face of model is compressed and can be later rendered with tessellation shaders (decompression is done during rasterization). Also, we propose a way of adding streaming support to our compression method to further reduce memory consumption. Finally, we made a comparison with state of the art approach levels of detail (LODs) approach and found that proposed approach has much lower memory consumption without negative effects to rasterization performance and quality