PacketPatch/
├── README.md                          # Project overview (this file)
├── LICENSE                            # MIT License
├── requirements.txt                   # Python dependencies
├── bert_base_config.json              # BERT model configuration (12 layers, 768-dim, 12 heads)
│
├── packet_adv/                        # ★ Core experimental code
│   ├── README.md                      # Usage guide for core modules
│   ├── adv_fine_tuning_cls.py         # Step 1: Train PatchGenerator (SpanBERT)
│   ├── generate.py                    # Step 2: Generate adversarial packets
│   └── gen_onnx.py                    # Step 3 (optional): ONNX/TensorRT acceleration
│
├── uer/                               # UER (Universal Encoder Representations) framework
│   ├── README.md                      # Framework documentation
│   ├── encoders/                      # Encoder implementations (Transformer, RNN, CNN)
│   ├── layers/                        # Network layers (Embedding, Attention, FFN)
│   ├── targets/                       # Training targets (MLM, NSP, Span-MBM, etc.)
│   ├── models/                        # Model definition
│   ├── utils/                         # Utilities (tokenizers, optimizers, data loading)
│   ├── model_builder.py               # Model construction
│   ├── model_loader.py                # Pretrained weight loading
│   ├── model_saver.py                 # Model checkpoint saving
│   ├── trainer.py                     # Training loop implementation
│   └── opts.py                        # CLI argument definitions
│
├── dataset/                           # Dataset documentation
│   └── README.md                      # Dataset descriptions, download links, preprocessing
│
├── image/                             # Architecture and result figures
│   └── README.md                      # Figure descriptions
│
└── tools/                             # Utility tools
    └── README.md                      # Tool descriptions and usage

# Clone this repository
git clone https://github.com/your-org/PacketPatch.git
cd PacketPatch

# Install dependencies
pip install -r requirements.txt

cd packet_adv

# Edit adv_fine_tuning_cls.py first:
#   - Update dataset paths (lines 47-50)
#   - Update pretrained model path (lines 73-75)
#   - Set correct num_classes (line 56)
#   - Set correct training set size (line 878)

python adv_fine_tuning_cls.py

cd packet_adv

# Edit generate.py first:
#   - Update test data paths (lines 42-44)
#   - Update --load_model_path to your trained checkpoint (line 68-69)
#   - Set BWO (bandwidth overhead ratio, line 60, default 0.1)
#   - Choose generation strategy via 'fun' parameter (line 50)

python generate.py

cd packet_adv

# Edit gen_onnx.py first:
#   - Update model checkpoint path (line 132)
#   - Update output paths (lines 145, 207)
#   - Select precision mode: 'fp16' or 'int8' (line 21)

python gen_onnx.py

@article{xu2025packetpatch,
    title = {PacketPatch: Practical Generation and Deployment of Adversarial Packets for Byte-Feature-Based Encrypted Traffic Classification},
    author = {Yuwei Xu and Yuanyuan Xu and Yunpeng Bai and Jiahui Chen and Kehui Song and Jie Cao and Qiao Xiang and Guang Cheng},
    journal = {Elsevier},
    year = {2025},
}