Generative Adversarial Networks for Extreme Learned Image Compression

TL;DR

This paper introduces a GAN-based learned image compression system that achieves high-quality, visually pleasing results at extremely low bitrates, outperforming existing methods and enabling semantic-based synthesis of unimportant regions.

Contribution

The paper proposes a novel GAN-based framework for low-bitrate image compression that synthesizes missing details and allows semantic-guided image synthesis, reducing storage costs.

Findings

Outperforms state-of-the-art methods at low bitrates in user studies

Synthesizes realistic details where data is limited

Enables semantic label-based synthesis of unimportant regions

Abstract

We present a learned image compression system based on GANs, operating at extremely low bitrates. Our proposed framework combines an encoder, decoder/generator and a multi-scale discriminator, which we train jointly for a generative learned compression objective. The model synthesizes details it cannot afford to store, obtaining visually pleasing results at bitrates where previous methods fail and show strong artifacts. Furthermore, if a semantic label map of the original image is available, our method can fully synthesize unimportant regions in the decoded image such as streets and trees from the label map, proportionally reducing the storage cost. A user study confirms that for low bitrates, our approach is preferred to state-of-the-art methods, even when they use more than double the bits.