Code128-Net

A deep learning-based system for detecting and localizing Code128 finder patterns from normalized 1D signals.

The project includes:

• 📦 Dataset loading and preprocessing
• 🧠 Model definition and training (PyTorch)
• 🔍 Evaluation and plotting tools
• 🛠️ CLI scripts for training, prediction, and processing

---

🔧 Installation

git clone https://github.com/aidinattar/code128-net.git
cd code128-net
make requirements

You can also create a virtual environment with make create_environment if you use Conda.

---

🚀 Quickstart

Train a model:

make train

Run inference on test data:

make predict

Evaluate results and generate plots:

make evaluate

Run all tests:

make test

---

🧠 Project Structure

code128-net/
├── data/               # Raw, processed and interim datasets
├── models/             # Trained model weights
├── reports/            # Output plots and evaluation results
├── code128net/         # Main source code
│   ├── config.py
│   ├── dataset.py
│   ├── features.py
│   ├── plots.py
│   ├── evaluation.py
│   └── modeling/
│       ├── model.py
│       ├── train.py
│       └── predict.py
├── tests/              # Unit tests with pytest
├── requirements.txt
├── Makefile
└── README.md

---

📈 Model Outputs

The model predicts 4 outputs for each signal:

• class: the type of finder pattern (StartA, StartB, StartC, Stop, NoPattern)
• index: the starting index of the pattern
• ppe: the pattern's proportional position estimate
• direction: the direction (0 or 1)

---

🧪 Testing

Unit tests are implemented using pytest:

pytest tests/

Tests cover:

• Model architecture and forward pass
• Dataset and data loading
• Loss computation and training loop
• CLI scripts (via typer.testing.CliRunner)

---

📄 License

This project is open-source and available under the MIT License.

---

✍️ Author

Created by Aidin Attar for a deep learning assignment on 1D signal analysis using PyTorch.