🧠 MNIST Digit Classifier (PyTorch CNN)

A deep learning project implementing a Convolutional Neural Network (CNN) for handwritten digit classification on the MNIST dataset using PyTorch.

---

🚀 Features

• ✅ CNN-based architecture (Conv + Pool + FC)
• ✅ Dropout for regularization
• ✅ Normalized input pipeline
• ✅ Modular project structure (production-ready)
• ✅ Cross-platform execution (Windows + Linux)
• ✅ ~99.26% test accuracy
• ✅ Visualization of predictions

---

🧱 Model Architecture

Architecture Details:

• Input: 1 × 28 × 28
• Conv1: 32 filters, 3×3 → ReLU → MaxPool → 32 × 14 × 14
• Conv2: 64 filters, 3×3 → ReLU → MaxPool → 64 × 7 × 7
• Flatten → 3136
• FC1: 128 neurons → ReLU → Dropout (0.25)
• FC2: 10 output classes

---

📊 Results

• Test Accuracy: 99.26%
• Final Training Loss: ~0.026
• Stable convergence within 5 epochs

---

📁 Project Structure

mnist-digit-classifier/
│
├── src/
│   ├── data/        # Data loading & preprocessing
│   ├── models/      # CNN architecture
│   ├── training/    # Training loop
│   ├── evaluation/  # Evaluation logic
│   └── utils/       # Visualization
│
├── configs/         # Hyperparameters
├── scripts/         # Run scripts
├── outputs/         # Saved models
├── notebooks/       # Jupyter experiments

---

⚙️ Installation

git clone <your-repo-url>
cd mnist-digit-classifier
pip install -r requirements.txt

---

🏋️ Training

✅ Recommended (cross-platform)

python -m src.training.train

🐧 Linux / Mac

bash scripts/train.sh

🪟 Windows

python src/training/train.py

---

📈 Evaluation

python -m src.evaluation.evaluate

Output:

Test Accuracy: 99.26%

---

🖼️ Sample Predictions

The model:

• Correctly classifies handwritten digits
• Generalizes well to different writing styles
• Achieves high confidence predictions

---

🧠 Tech Stack

• Python
• PyTorch
• Torchvision
• Matplotlib

---

🔮 Future Improvements

• 📊 Confusion Matrix & Precision/Recall
• 🔥 Grad-CAM visualization
• 📈 Training curves (loss/accuracy plots)
• 🧪 Hyperparameter tuning
• 🌐 Deploy as web app (Streamlit)

---

📌 Notes

• Dataset auto-downloads (no manual setup needed)
• GPU used automatically if available
• Modular design → easy to extend and scale

---

⭐ Acknowledgements

• MNIST Dataset
• PyTorch Documentation

---

💡 Author

Gopal Gupta

---

If you found this useful, consider giving it a ⭐ on GitHub!