Abstract
This document describes systems and methods for asymmetric video streaming that can address challenges associated with delivering high-resolution content over limited-bandwidth networks. A server-side system can analyze a high-resolution source video to generate a low-resolution base video stream and a separate metadata stream containing content-specific generative instructions. A client device, for example, a smartphone, smart television, or other computing device, can receive both streams and use the metadata to guide an on-device generative process, for instance, on a neural processing unit, to reconstruct a high-resolution video. This technique may reduce the amount of data transmitted, which can lower server costs and improve playback stability, while using source-specific information may improve the visual quality of a resulting output compared to some generic upscaling algorithms.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Garg, Rohan, "Asymmetric Video Streaming with Generative Metadata for Client-Side Reconstruction", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10848