FAST: Free Adaptive Super-Resolution via Transfer for Compressed Videos

High resolution displays are increasingly popular, requiring most of the existing video content to be adapted to higher resolution. State-of-the-art super-resolution algorithms mainly address the visual quality of the output instead of real-time throughput. This paper introduces FAST, a framework to accelerate any image based super-resolution algorithm running on compressed videos. FAST leverages the similarity between adjacent frames in a video. Given the output of a superresolution algorithm on one frame, the technique adaptively transfers it to the adjacent frames and skips running the super-resolution algorithm. The transferring process has negligible computation cost because the required information, including motion vectors, block size, and prediction residual, are embedded in the compressed video for free. In this work, we show that FAST accelerates state-of-the-art super-resolution algorithms by up to an order of magnitude with acceptable quality loss of up to 0.2 dB. Thus, we believe that the FAST framework is an important step towards enabling real-time super-resolution algorithms that upsample streamed videos for large displays.