Variational Image Compression with a Scale Hyperprior

Year: Feb 2018

Authors: Johannes Balle, David Minnen, Saurabh Singh, Sung Jin Hwang, Nick Johnston

Affiliations: Google

The authors describe an end-to-end trainable model for image compression based on variational autoencoders. The model incoporates a hyperprior to effectively capture spatial dependencies in the latent representation.

This model leads to SOTA image compression when measuring visual quality with MS-SSIM, and yields better rate-distortion performance than previous methods when evaluated on PSNR.