End-to-end optimized image compression with competition of prior distributions

TL;DR

This paper introduces a novel image compression method using a single convolutional autoencoder with multiple competing learned priors, achieving comparable rate-distortion performance with reduced complexity.

Contribution

It proposes a competition-based prior distribution approach for entropy coding, simplifying the process compared to traditional autoencoder-based methods.

Findings

Achieves rate-distortion performance comparable to predicted priors.

Reduces entropy coding and decoding complexity.

Uses a static table of learned priors for inference.

Abstract

Convolutional autoencoders are now at the forefront of image compression research. To improve their entropy coding, encoder output is typically analyzed with a second autoencoder to generate per-variable parametrized prior probability distributions. We instead propose a compression scheme that uses a single convolutional autoencoder and multiple learned prior distributions working as a competition of experts. Trained prior distributions are stored in a static table of cumulative distribution functions. During inference, this table is used by an entropy coder as a look-up-table to determine the best prior for each spatial location. Our method offers rate-distortion performance comparable to that obtained with a predicted parametrized prior with only a fraction of its entropy coding and decoding complexity.