ASRSR: Adaptive Sending Resolution and Super-resolution for Real-time Video Streaming

Real-time video streaming application has been adopted for a wide range of services in recent years. A major challenge for real-time video streaming is the low resolution and high latency caused by limited and unstable network bandwidth. A straightforward solution is to invest directly in network infrastructure, but it is cost inefficient and still limited to the simple transmitter of the video sender. To address these challenges, we are motivated to develop an alternative solution by leveraging video super-resolution (VSR). We propose a new adaptive sending resolution and super-resolution (ASRSR) scheme for real-time video streaming. ASRSR jointly decides the sending resolution for the sender and the super-resolved resolution for the VSR model at the receiver, according to changing network conditions to simultaneously optimize bandwidth demand and video quality. We evaluate the ASRSR system in a trace-driven simulation environment, demonstrating ASRSR system outperforms all benchmarks, and both the VSR and resolution adaptation algorithm contributes to significant performance gain.