CTDP Depacking with Guided Depth Upsampling Networks for Realization of Multiview 3D Video

With the advancement of 3D display technology, the original 3D stereo-view systems requiring 3D glasses have evolved to multiple-view naked-eyes 3D displays. For 2D broadcasting systems, we can pack one view and its corresponding depth map into a frame in the transmission side and generate multiple views by the depth image-based rendering (DIBR) engine at the receivers. Up to now, the centralized texture depth packing (CTDP) frame-compatible format is the best solution for view and depth packing. In this paper, we propose an improved CTDP depacking method and realize it with DIBR process to achieve real-time 3D multiview exhibition. To maintain the quality of the depacked depth map, in the CTDP depacking process, we suggest a deep learning-based guided depth upsampling network. The experimental results demonstrate that the proposed 3D multiview system can successfully depack CTDP video and generate 9-view 3D movies in real-time. The proposed guided depth upsampling network achieves a better quality of the generated views than the traditional algorithms.