Efficient Learned Image Compression without Entropy Coding

2026-05-22

AI Summary

This work addresses the high encoding and decoding latency in conventional learned image compression methods, which rely on entropy coding. The authors propose EF-LIC, a multi-rate framework that achieves entropy-coding-free learned compression for the first time. By integrating unconstrained vector quantization with a context-conditioned autoregressive transform, the method effectively eliminates both statistical and spatial redundancies. Leveraging maximum-entropy distribution analysis, the paper theoretically demonstrates that EF-LIC’s redundancy removal capability matches that of traditional entropy-coded approaches. Experimental results on the Kodak dataset show that, compared to MS-ILLM, EF-LIC reduces bitrate by 67.86% under the LPIPS metric while accelerating encoding and decoding speeds by 3× and 5×, respectively, achieving compression performance on par with state-of-the-art entropy-coded schemes.

Abstract

Entropy coding is widely used in typical learned image compression (LIC) that converts latents into a compact bitstream. However, entropy coding is typically sequential and becomes the coding latency bottleneck. To overcome it, we present Entropy-Coding Free Learned Image Compression (EF-LIC), a multi-rate framework that generates compact representation by removing statistical and correlation redundancy with low coding latency. First, we introduce unconstrained vector quantization and prove that its index distribution approaches the maximum-entropy bound, yielding minimal statistical redundancy. Second, we propose a context-conditioned autoregressive transform that directly reparameterizes the latents to reduce inter-dependency. Theoretical analysis shows that EF-LIC can remove correlation redundancy as effectively as typical LIC with entropy coding, leading to comparable compression performance. Experiments show EF-LIC achieves up to 67.86% bitrate reduction over MS-ILLM on Kodak with LPIPS. Ablation studies further show EF-LIC matches the compression performance of its entropy-coding based variant while achieving over $3\times$ faster encoding and $5\times$ faster decoding.

Problem

• learned image compression
• entropy coding
• coding latency
• statistical redundancy
• correlation redundancy

Innovation

• learned image compression
• entropy coding free
• vector quantization
• autoregressive transform
• low-latency compression