PROJECT TITLE :

Supremo Cloud-Assisted Low-Latency Super-Resolution in Mobile Devices

ABSTRACT:

We present Supremo, an image super-resolution (SR) system for low-latency use in mobile devices that is assisted by the cloud. Because SR requires a significant amount of computing power, we began by improving upon the state-of-the-art DNN in order to decrease the inference latency. In addition, we devise a mobile-cloud cooperative execution pipeline that is made up of specialized data compression algorithms. The goal of this pipeline is to reduce end-to-end latency while maintaining a high level of image quality. Finally, we extend Supremo to video applications by developing a dynamic optimal control algorithm for the design of Supremo-Opt. This algorithm's goal is to maximize the impact of SR while simultaneously satisfying latency and resource constraints under realistic network conditions. Supremo is able to upscale a 360p image to 1080p in 122 milliseconds, which is 43.68 times faster than the GPU execution on the device itself. Supremo reduces wireless network bandwidth consumption and end-to-end latency by 15.23% and 4.85%, respectively, when compared to cloud offloading-based solutions. Additionally, Supremo achieves 2.39 dB higher PSNR when compared to using conventional JPEG to achieve similar levels of data size compression. In addition, Supremo-Opt ensures a reliable performance in real-world situations.