中文无需熵编码的高效学习型图像压缩

提出无熵编码学习图像压缩（EF-LIC），一种多速率框架，通过无约束向量量化去除统计与相关性冗余，证明其索引分布逼近最大熵，从而降低编码延迟。

arXiv:2605.23323v1 Announce Type: cross 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-ent