🌀 Diffusion-Based Character Image Generator

A PyTorch-based implementation of a Denoising Diffusion Probabilistic Model (DDPM) trained on the EMNIST dataset to generate handwritten characters from noise.

---

📦 Dataset

I used the EMNIST dataset to train a character generation model. It includes thousands of grayscale handwritten character images.

from torchvision.datasets import EMNIST

---

🚀 How to Run

1. Clone the repo and navigate to the notebook:

git clone https://github.com/marounilabuni/diffusion-handwrite.git
cd diffusion-handwrite

2. Install dependencies:

pip install torch torchvision matplotlib tqdm

3. Open the notebook:

jupyter notebook diffusion_images_final.ipynb

---

🧪 Sampling Preview

Once trained, the model can generate samples like these:

---

📁 Project Structure

├── diffusion_images_final.ipynb   # Main training & sampling
└── README.md                      # Project overview

---

🧠 Future Work

• ✅ Add DDIM sampling
• 🔄 Add case-sensitive character generation (upper vs lower)
• 🧬 Enable style-conditioned character generation to mimic specific handwriting

---

📄 License

This project is licensed under the Apache 2.0 License. See the LICENSE file for details.

---

Crafted with 💻 + ☕ by Maroun Ilabuni.