# Real-Time Adaptive Image Compression

**Authors:** Oren Rippel, Lubomir Bourdev  
**arXiv:** [1705.05823](https://arxiv.org/abs/1705.05823)  
**Date:** 2017-05-16  
**Summary page:** https://www.summarizepaper.com/en/paper/1705.05823v1/

## Key points

- Oren Rippel and Lubomir Bourdev present a machine learning-based approach to lossy image compression that outperforms all existing codecs while running in real-time.
- The algorithm typically produces files 2.5 times smaller than JPEG and JPEG 2000, 2 times smaller than WebP, and 1.7 times smaller than BPG on datasets of generic images across all quality levels.
- The codec is designed to be lightweight and deployable, with the ability to encode or decode the Kodak dataset in around 10ms per image on GPU.
- The architecture is an autoencoder featuring pyramidal analysis, an adaptive coding module, and regularization of the expected codelength.
- The approach is supplemented with adversarial training specialized towards use in a compression setting, enabling visually pleasing reconstructions for very low bitrates.
- This new approach represents a significant improvement over existing codecs both in terms of compression ratio and speed of operation.
- Potential applications include video streaming where fast compression algorithms are essential for delivering high-quality content over limited bandwidth connections.

## Summary

In their paper "Real-Time Adaptive Image Compression," authors Oren Rippel and Lubomir Bourdev present a machine learning-based approach to lossy image compression that outperforms all existing codecs while running in real-time. The algorithm typically produces files 2.5 times smaller than JPEG and JPEG 2000, 2 times smaller than WebP, and 1.7 times smaller than BPG on datasets of generic images across all quality levels. The codec is designed to be lightweight and deployable, with the ability to encode or decode the Kodak dataset in around 10ms per image on GPU. The architecture is an autoencoder featuring pyramidal analysis, an adaptive coding module, and regularization of the expected codelength. Additionally, the approach is supplemented with adversarial training specialized towards use in a compression setting, enabling visually pleasing reconstructions for very low bitrates. This new approach represents a significant improvement over existing codecs both in terms of compression ratio and speed of operation. The authors note that their method has potential applications in fields such as video streaming where fast compression algorithms are essential for delivering high-quality content over limited bandwidth connections. Overall, this paper presents a promising new direction for lossy image compression using machine learning techniques that could have far-reaching implications for a variety of industries.