Efficient Context-Aware Lossy Image Compression

Jan Xu, Alex Lytchier, Ciro Cursio, Dimitrios Kollias, Chri Besenbruch, Arsalan Zafar

Abstract: We present an efficient context-aware lossy image compression system to participate in the Low Rate track of the CLIC 2020 Image Compression challenge. Our method is based on an auto-encoder pipeline augmented with a nested hyperprior model, a PixelCNN-based context model and an adversarial loss to remove artefacts.