帮我寻找一下,使用vq编解码进行图像压缩,利用数字系统传输index的语义通信的论文,最好是带有源码...

Создано: 11 апреля 2026 г.

Отвечено с использованием GPT-5.4 Thinking от Chat01

Вопрос

帮我寻找一下,使用vq编解码进行图像压缩,利用数字系统传输index的语义通信的论文,最好是带有源码,最近三年之内

Мысль

undefined

Мысль

undefined

Мысль

undefined

Мысль

undefined

Мысль

undefined

Мысль

undefined

Мысль

undefined

Мысль

undefined

Мысль

Clarifying code availability and citations

Since my search didn’t result in finding code, I could mention that as “I haven’t found a confirmed code repository” without requiring citation. I’ll phrase it like “源码状态:未在本轮公开检索中确认” to reflect my own search outcome, but after browsing, I’ll consider adding citations. Also, MOC-RVQ’s repo exists but isn’t released yet.

Organizing answer clearly with citations

I could mention the 2024 paper “Low-Rate Semantic Communication with Codebook-based Conditional Generative Models,” though its code status is uncertain. My goal is to keep answer concise, organized with headings and minimal categories, ensuring each bullet contains 1-2 lines and includes a citation. For follow-up, I can narrow down to specific categories if needed.

Thought for 2m 45s

Ответ

我先给你一个筛过一轮后的结论:

如果你的目标是“VQ/码本量化 + 发送离散 index + 图像语义通信”,近三年里最值得先看的,是下面 6 篇。
其中我把它们分成两类:已确认有公开代码,和方向非常贴但我这轮公开检索里没确认到可直接用的代码

一、已确认有公开代码

1) Improving Learning-Based Semantic Coding Efficiency for Image Transmission via Shared Semantic-Aware Codebook

年份:2025,IEEE TCOM
代码:有,官方 PyTorch 仓库
为什么推荐:这篇虽然不叫 VQ-VAE,但它非常贴你要的“图像语义压缩 + 码本/索引 + 数字系统传输”。核心是先构建 shared semantic-aware codebook,把每张图映射到 codeword,并把语义信息封装进 codeword indexes,再辅助端到端语义编解码。仓库里还直接给了 codebook construction 和训练脚本。(GitHub)

2) SQ-GAN: Semantic Image Communications Using Masked Vector Quantization

年份:2025,arXiv
代码:有,官方 GitHub
为什么推荐:这篇更偏“VQ 图像压缩”和“任务语义保真”。它用 masked VQ 做语义相关区域的选择性编码,在极低码率下兼顾感知质量和下游语义分割性能。仓库给了训练、评测、预训练模型下载。(arXiv)

3) Robust Semantic Communications With Masked VQ-VAE Enabled Codebook

年份:2023,IEEE TWC
代码:有,官方 GitHub
为什么推荐:这篇很贴“VQ 编解码 + 共享 codebook + 传输 index”。文中明确说发端只需要传输重要任务相关特征在 codebook 中的 indices;代码仓库也说明这是该 TWC 论文的 demo,并提供了训练/执行脚本。它比纯重建更偏“鲁棒语义通信/任务语义”。(arXiv)

二、非常贴题,但这轮没有确认到现成可跑代码

4) VQ-DeepISC: Vector Quantized-Enabled Digital Semantic Communication with Channel Adaptive Image Transmission

年份:2025,arXiv
为什么推荐:这是你需求里最“正中靶心”的一篇之一。它把语义特征投影到离散 latent/codebook,直接做 index-based transmission,并且明确落到 QPSK + OFDM + IEEE 802.11a 这种数字通信实现上。很适合看“如何把 VQ 语义表示真正接到数字链路上”。(arXiv)

5) VQ-DSC-R: Robust Vector Quantized-Enabled Digital Semantic Communication With OFDM Transmission

年份:2026,arXiv
为什么推荐:这是 VQ-DeepISC 之后更“工程化/鲁棒化”的版本。它同样是 shared semantic quantized codebook + efficient index transmission,并进一步加了 ANDVQ、CDM、三阶段训练和 OFDM 场景鲁棒优化。要看最新思路,这篇很值得跟。(arXiv)

6) Vector Quantized Semantic Communication System (VQ-DeepSC)

年份:2023,IEEE Wireless Communications Letters
为什么推荐:这是这一条线里比较早、也很经典的“VQ 语义通信 for image transmission”论文。它把多尺度语义特征量化到 semantic embedding spaces,使系统兼容数字通信。虽然比较短,但很适合当入口文献。(Qifan Fu)

三、补充一篇:很相关,但仓库还没真正放代码

7) MOC-RVQ: Multilevel Codebook-Assisted Digital Generative Semantic Communication

年份:2024,GLOBECOM 2024 / arXiv
代码状态:有官方仓库,但 README 明确写着 “codebase and pre-trained model weights will be released soon”
为什么值得看:它非常强调“数字语义通信”与“多级 codebook / RVQ / index 范围压缩”的兼容性问题,方向上和你的需求非常一致。(arXiv)


我帮你按“匹配你需求的程度”排个序

如果你现在就要开始读,我建议这个顺序:

  1. VQ-DeepISC (2025):最像“VQ + index + 数字传输”的标准答案。(arXiv)
  2. Shared Semantic-Aware Codebook (2025, TCOM):有完整代码,复现价值最高。(GitHub)
  3. VQ-DSC-R (2026):最新、最强工程味。(arXiv)
  4. SQ-GAN (2025):更偏图像压缩/感知质量,代码也比较完整。(arXiv)
  5. Robust Semantic Communications with Masked VQ-VAE Enabled Codebook (2023):更偏鲁棒与任务语义。(arXiv)
  6. VQ-DeepSC (2023):适合打基础。(Qifan Fu)

如果你只想要“带源码、尽量能复现”的,我会优先留这 3 篇

  • Shared Semantic-Aware Codebook (2025, TCOM) (GitHub)
  • SQ-GAN (2025) (arXiv)
  • Robust Semantic Communications With Masked VQ-VAE Enabled Codebook (2023) (arXiv)

一个小提醒

你这类需求里,文献可以分成两条路线:

  • 严格的“数字语义通信”路线:VQ 后传 index/bitstream,更接近通信系统实现。代表是 VQ-DeepSC / VQ-DeepISC / VQ-DSC-R / MOC-RVQ。(Qifan Fu)
  • “语义驱动图像压缩”路线:也用 VQ,但更强调低码率下的任务保真和生成质量。代表是 SQ-GAN。(arXiv)

你要是愿意,我下一条可以直接帮你整理成一份 “论文题目 + 链接 + 代码链接 + 核心方法 + 复现难度” 的精简表。

Поделиться этим Q&A