| Title |
On-Device Facial Light Field Synthesis System |
| Authors |
박준형(Junhyeong Bak) ; 박인규(In Kyu Park) |
| DOI |
https://doi.org/10.5573/ieie.2021.58.5.68 |
| Keywords |
Lightfield; multi-view synthesis; on-device; 3D modeling; Caffe |
| Abstract |
In this paper, we propose a light field synthesis system using a monocular facial image as input in an on-device environment. In order to overcome the difficulty of acquiring the ground truth facial light field in supervised learning, the proposed technique uses the synthesized light field acquired for various faces using a 3D modeling tool as learning data. In addition, a network compression method and mobile optimized deep learning framework are used for on-device implementation. The proposed system also includes viewpoint change and depth estimation functions that can be applied to the acquired light field image. For these, we apply OpenCL base GPU parallel processing so that these functions can be processed quickly in an on-device environment. Finally, this paper evaluates the operation results and processing time of our system implemented as GUI program on each platform. |