Wave domain llm improves lossless image compression accuracy

WaveLLM: Wavelet-Domain LLM Entropy Modeling for Lossless Image Compression

Computer Vision and Pattern Recognition

Summary

Lossless image compression aims to exactly shrink image size without losing any details, which requires very precise prediction models. The authors found that by breaking images into structured wavelet parts, their language model can predict these smaller pieces more easily than raw pixels. They designed WaveLLM, which uses wavelet decomposition and specialized interfaces to predict and compress images more accurately. Their method outperformed previous models on standard image datasets.

What this means in practice

  • For image codec developers: Build improved lossless image compressors with better accuracy using wavelet domain modeling combined with large language models.
  • For digital media storage providers: Deploy compressors that reduce storage size of images without quality loss by leveraging autoregressive wavelet-based LLM entropy coding.

Authors

Yue Yu, Bowei Kang, Changsheng Gao, Dong Liu, Xiaoyan Sun

Abstract

Lossless image compression fundamentally relies on accurate entropy modeling, driving a recent shift toward large language model (LLM)-based methods. However, existing approaches mainly emphasize the modeling capacity of LLMs while directly serializing images into pixel or residual sequences, leaving heterogeneous signal patterns entangled within a single prediction stream. In this paper, we argue that effective entropy modeling depends not only on model capacity, but also on the predictability of the input representation. Motivated by this insight, we propose WaveLLM, a wavelet-domain LLM-based entropy model for lossless image compression. By decomposing an image into multi-level wavelet subbands, WaveLLM converts a heterogeneous prediction problem into a set of more structured and statistically homogeneous subproblems that are more amenable to autoregressive probability estimation. To bridge numerical coding symbols with the LLM's native text token space, we introduce a dedicated numerical interface featuring value embeddings and a coefficient prediction head. Furthermore, a task-adaptive prompting strategy is designed to enable a single shared model to adapt to diverse subbands within a unified framework. Built on Llama-3.2-3B with LoRA adaptation, WaveLLM achieves 2.51, 2.24, and 2.27 bpsp on Kodak, CLIC.pro, and DIV2K, respectively, outperforming the PixelCNN-based codec using the same wavelet transform by 0.43, 0.36, and 0.46 bpsp, and surpassing recent LLM-based lossless compressors on all three benchmarks.