Advances in Diffusion-Based Generative Compression
It addresses the problem of efficient image compression for applications in visual media, but it is incremental as it provides a review and unification of existing approaches.
The paper reviews diffusion-based methods for generative lossy compression, focusing on image compression, where realistic reconstructions are generated at low bit-rates by encoding sources into embeddings and using diffusion models for iterative refinement.
Popularized by their strong image generation performance, diffusion and related methods for generative modeling have found widespread success in visual media applications. In particular, diffusion methods have enabled new approaches to data compression, where realistic reconstructions can be generated at extremely low bit-rates. This article provides a unifying review of recent diffusion-based methods for generative lossy compression, with a focus on image compression. These methods generally encode the source into an embedding and employ a diffusion model to iteratively refine it in the decoding procedure, such that the final reconstruction approximately follows the ground truth data distribution. The embedding can take various forms and is typically transmitted via an auxiliary entropy model, and recent methods also explore the use of diffusion models themselves for information transmission via channel simulation. We review representative approaches through the lens of rate-distortion-perception theory, highlighting the role of common randomness and connections to inverse problems, and identify open challenges.