Training-Free Coverless Multi-Image Steganography with Access Control

1. Environment Setup

# Create and activate conda environment
conda create -n MIDAS python=3.12.9 -y
conda activate MIDAS

# Install PyTorch with CUDA 12.1 support
pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 \
    --index-url https://download.pytorch.org/whl/cu121

# Install other dependencies
pip install -r requirements.txt

---

2. Pretrained Models

Download the IP-Adapter and encoder models and place them in the ./models/ directory.

• IP-Adapter:
  • https://huggingface.co/h94/IP-Adapter/blob/main/models/ip-adapter-plus_sd15.bin
• Image Encoder:
  • https://huggingface.co/h94/IP-Adapter/blob/main/models/image_encoder/pytorch_model.bin

---

3. Dataset Preparation

Download the Stego260 dataset from:

• https://github.com/yujiwen/CRoSS

The expected directory structure is as follows:

MIDAS/
├── dataset
│   └── steganodataset_v1
│       ├── data
│       └── config.yaml
├── metrics
│   ├── metrics_degrad.py
│   ├── metrics_multi.py
│   └── metrics_single.py
├── models
│   ├── config.json
│   ├── ip-adapter-plus_sd15.bin
│   └── pytorch_model.bin
├── results
├── cover_generation.py
├── MIDAS_degrad.py
├── MIDAS_multi.py
├── MIDAS_single.py
├── README.md
├── requirements.txt
├── scheduling.py
└── ...

---

4. Execution Guide

Note on File Paths

Before running the scripts, ensure that the following arguments match your local environment:

• --yaml_path: Path to your dataset configuration file
  (default: ./dataset/steganodataset_v1/config.yaml)

• --save_path: Directory where the generated results will be stored.
  Example: ./results/MIDAS/stego260/2/seed9000/

---

A. Single Image Hiding (1 Image)

MIDAS Mode

python ./MIDAS_single.py \
    --pw 9000 \
    --mode 'MIDAS' \
    --key_type 'random_basis' \
    --personal_key_strength 0.5 \
    --private_strength 0.4 \
    --edit_strength 0.7 \
    --ref_strength 0.4 \
    --alpha 0.95

DiffStega* Mode (Baseline)

python ./MIDAS_single.py \
    --pw 9000 \
    --mode 'DiffStega*' \
    --key_type 'noise_flip' \
    --personal_key_strength 0.05 \
    --private_strength 0.6 \
    --edit_strength 0.6

---

B. Multi-Image Hiding (2 Images)

MIDAS Mode

python ./MIDAS_multi.py \
    --pw 9000 \
    --num_secret_image 2 \
    --mode 'MIDAS' \
    --key_type 'random_basis' \
    --personal_key_strength 0.4 \
    --key_strength 0.5 \
    --private_strength 0.4 \
    --edit_strength 0.7 \
    --ref_strength 0.4 \
    --alpha 0.95

DiffStega* Mode (Baseline)

python ./MIDAS_multi.py \
    --pw 9000 \
    --num_secret_image 2 \
    --mode 'DiffStega*' \
    --key_type 'noise_flip' \
    --personal_key_strength 0.05 \
    --key_strength 0.05 \
    --private_strength 0.6 \
    --edit_strength 0.6

Tip:
You can evaluate performance across various scales by varying --num_secret_image to 2, 4, or 8.

---

C. Degradation Test

Note:
The degradation test is currently implemented specifically for the 2-image hiding scenario. Ensure that you have completed the experiments in Section B before running this script.

MIDAS Mode

python ./MIDAS_degrad.py \
    --pw 9000 \
    --mode 'MIDAS' \
    --key_type 'random_basis' \
    --personal_key_strength 0.4 \
    --key_strength 0.5 \
    --private_strength 0.5 \
    --edit_strength 0.7 \
    --ref_strength 0.4 \
    --alpha 0.95

DiffStega* Mode (Baseline)

python ./MIDAS_degrad.py \
    --pw 9000 \
    --mode 'DiffStega*' \
    --key_type 'noise_flip' \
    --personal_key_strength 0.05 \
    --key_strength 0.05 \
    --private_strength 0.7 \
    --edit_strength 0.6

---

5. Evaluation & Analysis

A. Quantitative Evaluation

The scripts in the ./metrics directory provide comprehensive quality assessments.

B. Steganalysis Preparation

Use cover_generation.py to generate original cover images for comparative steganalysis.

---

Acknowledgement

Part of the code is borrowed from DiffStega:

• https://github.com/evtricks/DiffStega