📚 MNIST Handwritten Digit Classifier

📖 Overview

This project predicts handwritten digit classes (0–9) using the MNIST dataset and a convolutional neural network (CNN) built with PyTorch. It demonstrates a full machine learning pipeline from data loading to inference, including:

• 🧠 CNN with stacked convolutional layers, Batch Normalization, Max Pooling, LeakyReLU activation, and Dropout
• ⚖️ Cross-Entropy Loss for multi-class classification
• 🚀 Adam optimizer for gradient updates
• 🔀 Mini-batch training with DataLoader
• 📊 Train/Validation/Test split for robust evaluation
• 📈 Live training & validation loss monitoring
• ✅ Softmax activation on the output for probability distribution across 10 classes
• 🎨 Interactive Gradio Interface for real-time image classification visualization

---

🖼️ Application Screenshot

Below is a preview of the Gradio Interface used for real-time tire classification:

---

🧩 Libraries

• PyTorch – model, training, and inference
• pandas – data handling
• matplotlib – loss visualization
• pickle – saving/loading normalization params and trained model
• Gradio — interactive web interface for real-time model demos

---

⚙️ Requirements

• Python 3.13+
• Recommended editor: VS Code

---

📦 Installation

• Clone the repository

git clone https://github.com/hurkanugur/MNIST-Digit-Classifier.git

• Navigate to the MNIST-Digit-Classifier directory

cd MNIST-Digit-Classifier

• Install dependencies

pip install -r requirements.txt

• Navigate to the MNIST-Digit-Classifier/src directory

cd src

---

🔧 Setup Python Environment in VS Code

1. View → Command Palette → Python: Create Environment
2. Choose Venv and your Python version
3. Select requirements.txt to install dependencies
4. Click OK

---

📂 Project Structure

assets/
└── app_screenshot.png                # Screenshot of the application
└── 1, 2, 3 ... 9.png                 # Digit samples

data/
└── MNIST                             # MNIST dataset (raw greyscale images and labels)

model/
└── mnist_digit_classifier.pth        # Trained model (after training)

src/
├── config.py                         # Paths, hyperparameters, split ratios
├── dataset.py                        # Data loading & preprocessing
├── device_manager.py                 # Selects and manages compute device
├── main_train.py                     # Training & model saving
├── main_inference.py                 # Inference pipeline
├── model.py                          # Neural network definition
├── visualize.py                      # Training/validation plots

requirements.txt                      # Python dependencies

---

📂 Model Architecture

Input (1×28×28)

Conv Block 1:
  → Conv2d(in_channels=1, out_channels=32, kernel_size=3, stride=1, padding=1, padding_mode="reflect")
  → BatchNorm2d(32)
  → ReLU
  → MaxPool2d(kernel_size=2, stride=2)
  → Dropout(0.25)

Conv Block 2:
  → Conv2d(in_channels=32, out_channels=64, kernel_size=3, stride=1, padding=1, padding_mode="reflect")
  → BatchNorm2d(64)
  → ReLU
  → MaxPool2d(kernel_size=2, stride=2)
  → Dropout(0.25)

Fully Connected:
  → Flatten
  → Linear(64×7×7, 128)
  → ReLU
  → BatchNorm1d(128)
  → Dropout(0.5)
  → Linear(128, 10)
  → Softmax(Output)

---

📂 Train the Model

python main_train.py

or

python3 main_train.py

---

📂 Run Predictions on Real Data

python main_inference.py

or

python3 main_inference.py