Scalable multiple GPU architecture for super multi-view synthesis using MVD

Abstract

This paper presents a scalable multiple GPU architecture for super multi-view (SMV) synthesis using the multi-view video plus depth (MVD) data. SMV synthesis is essential to generate 3D contents for the SMV 3D display with hundred views. SMV 3D display, recently released to support 108 viewpoints, shows the multiplexed result of small viewing interval. Hence, we should synthesize the intermediate views over a hundred for each pair of two cameras in multi-camera system. View synthesis of more than hundred high resolution images, however, needs massive data processing, which is linearly increased in proportion to the number of synthesized views. In this paper, we propose a real-time SMV synthesis method using multiple GPU. The scalability of GPU can be utilized to reduce the processing time of view synthesis without any changes of the kernel function. We evaluate the proposed method for synthesizing 180 intermediate views from 18 input HD images according to the number of GPUs. We show that 180 intermediate views can be synthesized in real-time using 4 GPUs.