Soft 2D-to-3D Delivery Using Deep Graph Neural Networks for Holographic-Type Communication

Takuya Fujihashi, Toshiaki Koike-Akino, Takashi WatanabeICASSP 2023 · OpenReview

Abstract

Holographic-type communication, i.e., three-dimensional (3D) content delivery, will be a crucial application for modern wireless and mobile networks. In this paper, we propose a novel soft delivery scheme to realize efficient 3D content delivery. Specifically, the proposed scheme sends a single 2D image over error-prone wireless channels using discrete cosine transform followed by near-analog modulation. At the receiver, a 2D-to-3D decoder based on graph neural networks (GNN) reconstructs the corresponding 3D point cloud and mesh from the received 2D image. We verify that the proposed soft 2D-to-3D delivery scheme can reconstruct clean 3D data gracefully from the soft-delivered 2D image even in the presence of fading and noise distortion. In addition, the proposed scheme can generate higher-quality 3D data compared with direct 3D content delivery schemes.
BibTeX
@inproceedings{fujihashi2023soft,
  author = {Takuya Fujihashi and Toshiaki Koike-Akino and Takashi Watanabe},
  title = {{Soft 2D-to-3D Delivery Using Deep Graph Neural Networks for Holographic-Type Communication}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  url = {https://openreview.net/forum?id=zmgqVyMQUal},
}

Loading related content...

Figure