面向生成式图像压缩率失真优化的可微矢量量化
Differentiable Vector Quantization for Rate-Distortion Optimization of Generative Image Compression
把码本分布做可微松弛，让熵损失直接塑造潜变量先验，从而在极低码率下用轻量架构实现端到端率失真优化；这套可微信道建模思路也可迁移到其他需要离散潜变量的视觉生成/压缩流程。
原文依据 a differentiable relaxation of the codebook distribution, allowing the entropy loss to directly shape the latent prior
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论在严格的存储与带宽约束下视觉数据的快速增长，使极低码率图像压缩日益重要。矢量量化（VQ）虽能提供较强的结构保真度，但现有方法因表征学习与熵建模之间的脱节，缺乏联合率失真（RD）优化的原则性机制。我们提出 RDVQ，一个统一框架，通过对码本分布的可微松弛，使基于 VQ 的压缩能够进行端到端 RD 优化，从而让熵损失直接塑造潜变量先验。我们进一步设计了一个自回归熵模型，支持精确的熵建模与测试时码率控制。大量实验表明，RDVQ 以轻量架构在极低码率下取得强劲性能，在参数量显著更少的情况下达到有竞争力或更优的感知质量。与 RDEIC 相比，RDVQ 在 DIV2K-val 上于 DISTS 指标下将码率最多降低 75.71%，于 LPIPS 指标下最多降低 37.63%。除实证收益外，RDVQ 还引入了 VQ 的熵约束形式，凸显了图像分词与压缩走向更统一视角的潜力。代码见 https://github.com/CVL-UESTC/RDVQ。
英文原文摘要 1281 字符
The rapid growth of visual data under stringent storage and bandwidth constraints makes extremely low-bitrate image compression increasingly important. While Vector Quantization (VQ) offers strong structural fidelity, existing methods lack a principled mechanism for joint rate-distortion (RD) optimization due to the disconnect between representation learning and entropy modeling. We propose RDVQ, a unified framework that enables end-to-end RD optimization for VQ-based compression via a differentiable relaxation of the codebook distribution, allowing the entropy loss to directly shape the latent prior. We further develop an autoregressive entropy model that supports accurate entropy modeling and test-time rate control. Extensive experiments demonstrate that RDVQ achieves strong performance at extremely low bitrates with a lightweight architecture, attaining competitive or superior perceptual quality with significantly fewer parameters. Compared with RDEIC, RDVQ reduces bitrate by up to 75.71% on DISTS and 37.63% on LPIPS on DIV2K-val. Beyond empirical gains, RDVQ introduces an entropy-constrained formulation of VQ, highlighting the potential for a more unified view of image tokenization and compression. The code is available at https://github.com/CVL-UESTC/RDVQ.