Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Efficient Learned Image Compression without Entropy Coding

About

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.

Hao Cao, Wenqi Guo, Zhijin Qin, Jungong Han• 2026

Related benchmarks

TaskDatasetResultRank
Image CompressionKodak--
58
Image CompressionKodak (test)
BD-Rate (LPIPS)-67.86
35
Image CompressionCLIC 2020
BD-rate (DISTS)-42.15
34
Image CompressionKodak
BD-Rate (DISTS)-70.61
25
Image CompressionTecnick (test)
BD-rate (LPIPS)-55.46
21
Image CompressionDIV2K (test)
BD-Rate (LPIPS)-62.33
20
Image CompressionDIV2K
BD-Rate (DISTS)-60.43
19
Image CompressionTecnick
BD-Rate (DISTS)-43.12
13
Image CompressionCLIC 2020 (test)
BD-Rate (LPIPS)-63.22
11
Showing 9 of 9 rows

Other info

Follow for update