Provable Secure Steganography Based on Adaptive Dynamic Sampling

# ADS Steganography

This repo provides a minimal implementation of Adaptive Dynamic Sampling (ADS) steganography.

You can run the pipeline via ADS.ipynb,including the encoding and decoding algorithms.

## Repository layout

<summary><b>Repository structure</b></summary>

```text

ADS-github/

├── ADS.ipynb

├── README.md

├── requirements.txt

├── bit_stream.txt (secret messages, only for reference)

├── instinwild_en_4000.json (prompts)

└── XHS.jsonl (prompts)

```

## Installation

Python 3.9+ recommended

```text

pip install -r requirements.txt

```

### Model

In this experiment, we download large language models that are available on Hugging Face(https://huggingface.co/).

Users also can load the model directly from Hugging Face using a model ID.

 

## Quick Start

1. Insstall dependencies.

2. Open the notebook: ADS.ipynb

3. Run cells in order to reproduce encoding/decoding.

## Secret message

We use a randomly generated binary bitstream as the secret message in ADS.ipynb. The provided bit_stream.txt is only for reference; users may also choose to load bit_stream.txt as the secret message, which yields the same functionality.

 

## Datasets (prompts)

    •   XHS.jsonl (Chinese prompts, JSON Lines)

    •   instinwild_en_4000.json (English prompts, JSON)

These are for quick checks; replace with your own prompts for experiments.

## Ethics

 Steganography is not a panacea that takes the risk of circumvention away.

 This repository is provided strictly for academic evaluation. Please comply with applicable laws, regulations, and platform policies.