@article{Li2026, 
author = {Yuchen Li and Xinyi Li and Tingjuan Lu and Jin Zhou and Shuang Li and XiaoLi Gong and Jin Zhang},
title = {Nexus: On-Device Lightweight Real-Time Super-Resolution System for 360◦ Videos},
year = {2026},
journal = {Tsinghua Science and Technology},
keywords = {360◦ video, lightweight super-resolution, viewport prediction, real-time},
url = {https://www.sciopen.com/article/10.26599/TST.2026.90100014},
doi = {10.26599/TST.2026.90100014},
abstract = {Delivering high-quality 360◦ video on resource-constrained head-mounted displays (HMDs) is challenging due to limited bandwidth and compute. We present Nexus, a lightweight, on-device real-time super-resolution system. Nexus transmits only low-quality frames and performs high-quality reconstruction on the HMDs for predicted viewports. We also design a viewport filling scheme to enhance the system’s robustness. Our viewport prediction algorithm achieves a mean angular error of 0.58, and our lightweight SR algorithm reduces inference time by 61% compared to state-of-the-art lightweight SR methods. Under a frame drop rate below 0.1%, compared to full-frame super-resolution on VR devices, Nexus supports resolutions up to 720p (up from 480p), fixed at 30 fps, and increases the maximum tractable bitrate from 10 Mbps to 25 Mbps.}
}