Due to the limited computing power and storage on mobile devices, 3D visualization on mobile devices can be a very challenging task. For large-scale models, a large portion of the rendering pipeline currently needs to be placed in the cloud. Usually, only the visual result--that is, the final images--will be transmitted to the client. However, traditional video streaming methods may suffer from small network bandwidth and network latency.
In this paper, the authors propose a remote rendering algorithm using multilayered impostors. Instead of transferring the final image pixels, impostors are generated on the server side and transferred to the client side for rendering. One benefit of using impostors is that they can be reused in many frames, thus minimizing the data transfer between the server and the client as the camera moves.
To compute the impostors, the authors propose an image-based mesh simplification method that uses depth information from the depth buffer. The algorithm first creates multi-layered cover maps by downsampling the depth buffer. It also adaptively creates multi-layered subdivision maps by identifying the depth complexity in different regions of the depth buffer. The cover maps and the subdivision maps are then combined into fragment maps, which are used to generate the simplified meshes. The authors also propose several sampling rules to detect discontinuity and avoid artifacts. Finally, a skybox prioritizing scheme is proposed for impostor regeneration and incremental updates. The results show that, compared to video streaming, the proposed method requires significantly less network bandwidth and effectively reduces the workload of the server.
The paper is well written, with a clear goal and a good result. The algorithm itself is very interesting. The paper’s only weakness is the lack of comparison to other mesh simplification methods for building impostors. Since image-based methods are constrained by the resolution of the depth buffer, non-image-based methods may provide better performance.