SV-3D BSC: Semantic Communications from Single View to 3D Reconstruction

With the exploration of the virtual world, 3D reconstruction has drawn increasing attention. To achieve largescale and low-latency virtual world interaction, data volume compression and visual experience are indispensable, but they tend to conflict in practice. Based on the above issues, we propose a semantic communication-based image transmission and 3D generation scheme (SV-3D BSC). First, we perform semantic segmentation on the image at the transmitter. In addition, at the receiver, we analyze the semantic symbols based on a common knowledge base and preliminarily restore the image through a generative adversarial network (GAN). Next, the restored image is further optimized for generating a Neural Radiance Fields (NeRF) representation. This scheme can greatly compress the data volume of virtual interaction and generate exquisite 3D models. Finally, experiments have shown that our scheme saves a lot of bandwidth and generates reasonable and colorful 3D models.