CQILAB/GenSC-6G-Segmentation

# GenSC-6G - Scalable Semantic Communication Framework and Dataset

This repository contains the **first semantic communication dataset and playground**, designed to be scalable, reproducible, and adaptable for a wide range of applications. The dataset and framework are tailored for semantic decoding, classification, and localization tasks in 6G applications, integrating generative AI and semantic communication. Implementation of **[GenSC-6G: A Prototype Testbed for Integrated Generative AI, Quantum, and Semantic Communication](https://arxiv.org/abs/2501.09918)**.

---
## Citation
The paper can be found at [arXiv](https://arxiv.org/abs/2501.09918).
If you use this dataset or framework in your research, please cite:

```bibtex
@article{gensc6g,
title={GenSC-6G: A Prototype Testbed for Integrated Generative AI, Quantum, and Semantic Communication},
author={Brian E. Arfeto and Shehbaz Tariq and Uman Khalid and Trung Q. Duong and Hyundong Shin},
year={2025},
eprint={2501.09918},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2501.09918},
}
```
---

## Features of the GenSC-6G Dataset

### 🔧 Adaptable SC Framework
A flexible prototype that supports modifications to baseline models, communication modules, and decoders, enabling customization for diverse communication needs.

### 🤖 Generative AI-Driven SC
The integration of generative AI for synthetic data generation, enriching the Knowledge Base (KB) and leveraging large language model (LLM) capabilities for enhanced semantic tasks.

### 📊 Noise-Augmented Dataset
A labeled dataset with injected noise, specifically optimized for semantic tasks such as target recognition, localization, and recovery. The dataset comprises 4,829 training and 1,320 testing instances across 15 classes of military and civilian vehicle types. It incorporates Additive White Gaussian Noise (AWGN) and Radio Frequency (RF) interference at varying Signal-to-Noise Ratios (SNRs) to evaluate model robustness under realistic channel conditions.

### 📥 Dataset Download and Overview

#### Main Dataset
**[Download the main dataset here](https://huggingface.co/datasets/CQILAB/GenSC-6G)**

#### Segmentation Dataset
**[Download the segmentation dataset here](https://huggingface.co/datasets/CQILAB/GenSC-6G-Segmentation)**

## Setup Instructions
Can be found in Official Repository: [CQILAB/GenSC-6G](https://github.com/CQILAB-Official/GenSC-6G)

## Reproducibility

### 🗃️ Dataset
Labeled dataset with ground-truth data, noise features, and extracted semantic features. Uploaded to **[HuggingFace🤗](https://huggingface.co/datasets/CQILAB/GenSC-6G)**
#### Dataset Columns and Descriptions
- **image**: Raw image data used for training and evaluation.
- **image_path**: Path to the corresponding image file.
- **classification_class**: Integer label corresponding to the classification category (0-15).
- **classification_{basemodel}_features**: Extracted feature embeddings from `{basemodel}`'s encoder, consisting of 1000 float32 tensors.
- **classification_awgn10dB_{basemodel}_features**: Feature embeddings extracted from `{basemodel}` encoder with Additive White Gaussian Noise (AWGN) at 10dB SNR.
- **classification_awgn30dB_{basemodel}_features**: Feature embeddings extracted from `{basemodel}` encoder with AWGN at 30dB SNR.
- **upsampling_{basemodel}_features**: Extracted feature embeddings for upsampling tasks using `{basemodel}` encoder, consisting of 1000 float32 tensors.
- **upsampling_awgn10dB_{basemodel}_features**: Upsampling features with AWGN at 10dB SNR for `{basemodel}`.
- **upsampling_awgn30dB_{basemodel}_features**: Upsampling features with AWGN at 30dB SNR for `{basemodel}`.

### 🏗️ Testbed
To experiment with real-world semantic communication, you can use the **GNURadio and HackRF**.
1. **Install Dependencies**:
- Install [GNU Radio](https://www.gnuradio.org/)
- Install HackRF tools: `sudo apt install hackrf`
2. **Configure Transceiver**:
- Transmitter config: `GNURadio/transmitter.grc`
- Outputs a **streaming binary file**
3. **Run Transmitter**:
- Open `GNURadio/transmitter.grc` in GNU Radio Companion
- Set SDR parameters (frequency, gain, bandwidth)
- Execute to start transmission
4. **Run Receiver**:
- Modify `GNURadio/receiver.grc` settings
- Run to capture and process signals
By following these steps, you can replicate real-world transmission experiments using the testbed and analyze its performance.

### 📊 Performance Metrics & Flexible Code
Can be found in Official Repository: [CQILAB/GenSC-6G](https://github.com/CQILAB-Official/GenSC-6G)

## Others

- Official Repository: [CQILAB/GenSC-6G](https://github.com/CQILAB-Official/GenSC-6G)
- Dataset: [HuggingFace](https://huggingface.co/datasets/CQILAB/GenSC-6G)

## License
```
MIT License
```

# GenSC-6G——可扩展语义通信（semantic communication）框架与数据集

本仓库包含**首个语义通信数据集与实验平台**，旨在实现可扩展、可复现与广泛适配性，以支撑各类应用场景。该数据集与框架面向6G应用中的语义解码、分类与定位任务打造，集成了生成式AI（Generative AI）与语义通信技术，对应论文为**《GenSC-6G：集成生成式AI、量子与语义通信的原型测试床》**（[GenSC-6G: A Prototype Testbed for Integrated Generative AI, Quantum, and Semantic Communication](https://arxiv.org/abs/2501.09918)）。

---
## 引用
该论文可在[arXiv](https://arxiv.org/abs/2501.09918)获取。若您在研究中使用本数据集或框架，请引用如下文献：

bibtex
@article{gensc6g,
title={GenSC-6G: A Prototype Testbed for Integrated Generative AI, Quantum, and Semantic Communication},
author={Brian E. Arfeto and Shehbaz Tariq and Uman Khalid and Trung Q. Duong and Hyundong Shin},
year={2025},
eprint={2501.09918},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2501.09918},
}

---

## GenSC-6G数据集的特性

### 🔧 可扩展语义通信框架
一款灵活的原型系统，支持对基线模型、通信模块与解码器进行修改，可针对多样化通信需求实现定制化开发。

### 🤖 生成式AI驱动的语义通信
通过集成生成式AI实现合成数据生成，以丰富知识库（Knowledge Base, KB），并借助大语言模型（Large Language Model, LLM）的能力优化各类语义任务。

### 📊 噪声增强数据集
一款带注入噪声的标注数据集，专门针对目标识别、定位与恢复等语义任务优化。该数据集包含15类军用与民用车辆类型的4829个训练样本与1320个测试样本。数据集引入了不同信噪比（Signal-to-Noise Ratio, SNR）下的加性高斯白噪声（Additive White Gaussian Noise, AWGN）与射频（Radio Frequency, RF）干扰，用于评估模型在真实信道条件下的鲁棒性。

### 📥 数据集下载与概览

#### 主数据集
**[点击此处下载主数据集](https://huggingface.co/datasets/CQILAB/GenSC-6G)**

#### 分割数据集
**[点击此处下载分割数据集](https://huggingface.co/datasets/CQILAB/GenSC-6G-Segmentation)**

## 设置说明
设置说明详见官方仓库：[CQILAB/GenSC-6G](https://github.com/CQILAB-Official/GenSC-6G)

## 可复现性

### 🗃️ 数据集
带真实标签、噪声特征与提取语义特征的标注数据集，已上传至**[HuggingFace🤗](https://huggingface.co/datasets/CQILAB/GenSC-6G)**
#### 数据集字段与说明
- **image**：用于训练与评估的原始图像数据。
- **image_path**：对应图像文件的存储路径。
- **classification_class**：对应分类类别的整数标签（取值范围0-15）。
- **classification_{basemodel}_features**：从`{basemodel}`编码器提取的特征嵌入，包含1000个float32张量。
- **classification_awgn10dB_{basemodel}_features**：在信噪比为10dB的加性高斯白噪声（AWGN）环境下，从`{basemodel}`编码器提取的特征嵌入。
- **classification_awgn30dB_{basemodel}_features**：在信噪比为30dB的加性高斯白噪声（AWGN）环境下，从`{basemodel}`编码器提取的特征嵌入。
- **upsampling_{basemodel}_features**：使用`{basemodel}`编码器提取的上采样任务特征嵌入，包含1000个float32张量。
- **upsampling_awgn10dB_{basemodel}_features**：在信噪比为10dB的加性高斯白噪声（AWGN）环境下，`{basemodel}`的上采样任务特征嵌入。
- **upsampling_awgn30dB_{basemodel}_features**：在信噪比为30dB的加性高斯白噪声（AWGN）环境下，`{basemodel}`的上采样任务特征嵌入。

### 🏗️ 测试床
若要开展真实场景下的语义通信实验，可使用**GNURadio与HackRF**设备。
1. **安装依赖项**：
- 安装[GNU Radio](https://www.gnuradio.org/)
- 安装HackRF工具：`sudo apt install hackrf`
2. **配置收发机**：
- 发射机配置文件：`GNURadio/transmitter.grc`
- 输出**流式二进制文件**
3. **运行发射机**：
- 在GNU Radio Companion中打开`GNURadio/transmitter.grc`
- 设置软件定义无线电（Software Defined Radio, SDR）参数（频率、增益、带宽）
- 执行脚本以启动传输
4. **运行接收机**：
- 修改`GNURadio/receiver.grc`的配置参数
- 运行脚本以捕获并处理信号
按照上述步骤，您即可基于该测试床复现真实传输实验并分析其性能表现。

### 📊 性能指标与灵活代码
性能指标与灵活代码详见官方仓库：[CQILAB/GenSC-6G](https://github.com/CQILAB-Official/GenSC-6G)

## 其他

- 官方仓库：[CQILAB/GenSC-6G](https://github.com/CQILAB-Official/GenSC-6G)
- 数据集：[HuggingFace](https://huggingface.co/datasets/CQILAB/GenSC-6G)

## 许可证

MIT许可证