WaveLLM: Wavelet-Domain LLM Entropy Modeling for Lossless Image Compression
WaveLLM proposes a wavelet-domain Large Language Model (LLM) approach for lossless image compression. It decomposes images into structured wavelet subbands to simplify prediction, then uses a shared LLM with task-adaptive prompting to model entropy across these subbands. This method achieves competitive or superior lossless compression rates on three benchmarks.