Learning Patterns of Latent Residual for Improving Video Compression

Abstract

We tackle the problem of reducing compression artifacts. Specifically, we focus on transmitting the residual from the original video, i.e. difference between a compressed video and its corresponding original/uncompressed one, together with the compressed video during video transmission. Our video compression pipeline is capable of diminishing the overall cost of transmitting the residual and simultaneously achieving comparable video quality with respect to a state-of-the-art baseline. We provide experimental results based on various compression methods on video datasets with great diversity, to substantiate the capacity of our pipeline in improving video compression.