ELIC: Efficient Learned Image Compression with Unevenly Grouped Space-Channel Contextual Adaptive Coding (2203.10886v2)

Abstract: Recently, learned image compression techniques have achieved remarkable performance, even surpassing the best manually designed lossy image coders. They are promising to be large-scale adopted. For the sake of practicality, a thorough investigation of the architecture design of learned image compression, regarding both compression performance and running speed, is essential. In this paper, we first propose uneven channel-conditional adaptive coding, motivated by the observation of energy compaction in learned image compression. Combining the proposed uneven grouping model with existing context models, we obtain a spatial-channel contextual adaptive model to improve the coding performance without damage to running speed. Then we study the structure of the main transform and propose an efficient model, ELIC, to achieve state-of-the-art speed and compression ability. With superior performance, the proposed model also supports extremely fast preview decoding and progressive decoding, which makes the coming application of learning-based image compression more promising.

• The paper presents an innovative approach with uneven channel grouping that leverages energy compaction for faster decoding without sacrificing quality.
• It integrates a combined spatial-channel contextual model that capitalizes on redundancies in both dimensions to enhance compression efficiency.
• Empirical results on Kodak and CLIC benchmarks demonstrate lower bit rates and improved PSNR and MS-SSIM, confirming its practical advantages.

Overview of ELIC: Efficient Learned Image Compression with Unevenly Grouped Space-Channel Contextual Adaptive Coding

Introduction

Image compression has evolved significantly with the advent of learned techniques, which often outperform traditional methods such as JPEG and BPG in terms of rate-distortion performance. The paper presents a novel method, ELIC (Efficient Learned Image Compression) with Unevenly Grouped Space-Channel Contextual Adaptive Coding, which targets not only improved compression efficiency but also superior computational performance. By focusing on an uneven channel grouping strategy and integrating spatial-channel context modeling, the paper aims to address conventional limitations related to decoding speed, making it viable for practical implementation.

Methodology

The paper centers on several key innovative strategies:

1. Uneven Channel Grouping: Expanding on channel-conditional models, the authors introduce an uneven grouping approach that leverages the energy compaction property observed in learned image compression. Instead of uniformly distributing channels across groups, the method uses fewer channels in the early groups where most of the energy is concentrated. This results in faster processing without sacrificing performance.
2. Space-Channel Contextual Adaptive Model (SCCTX): By combining spatial and channel-wise context models, the SCCTX model captures redundancies in both dimensions, enhancing compression efficiency. The spatial context is modeled in a parallel manner, which does not impede the decoding speed significantly.
3. Nonlinear Transform via Residual Blocks: Instead of the commonly used GDN layers, the ELIC model employs stacked residual blocks to increase the network's nonlinearity. This choice allows for better compression performance due to improved feature representation, while also offering the potential for dynamic and scalable deployment.
4. Thumbnail Preview: The paper addresses practical applications such as quick thumbnail and progressive decoding by introducing a thumbnail synthesizer. This lightweight network facilitates very fast preview decoding, aiding applications where rapid content assessment is required.

Results

The ELIC framework was evaluated against several state-of-the-art models and showed superior performance both in terms of compression efficacy and computational efficiency. When tested on benchmarks like Kodak and CLIC, ELIC demonstrated reduced bit rates and decoding times compared to conventional models like VVC, while maintaining competitive or superior image quality (assessed via PSNR and MS-SSIM metrics).

Implications and Future Directions

The paper's contributions are significant for both theoretical and practical dimensions of image compression. By addressing speed and efficiency, the method is ideally positioned for real-world applications that require rapid and resource-efficient processing, such as image streaming and mobile photography.

Moving forward, future developments could include exploring the theoretical underpinnings of the information compaction property, which could further refine the uneven grouping strategy. Additionally, extending the model to other visual media types and integrating perceptual quality enhancements could broaden ELIC’s applicability. Integrating such models with fast hardware acceleration techniques could further optimize speed and enable real-time applications.

Overall, ELIC represents a substantial step forward in balancing compression performance and computational efficiency, paving the way for more widespread adoption of learned compression models in industry.