MNIST CNN 🧠

A simple Convolutional Neural Network built in TensorFlow to classify handwritten digits from the MNIST dataset.

• main branch → runs on CPU
• cuda branch → runs on GPU

---

CUDA Branch Only:

You must have CUDA and cuDNN installed manually (not included in requirements.txt).

If you have an older GPU, you may encounter TensorFlow bugs. In that case, try:

pip install tensorflow==2.10.0

---

🚀 Overview

This project demonstrates how to construct, train, and evaluate a CNN for recognizing digits (0–9) in grayscale 28×28 pixel images. The MNIST dataset is one of the foundational benchmarks in deep learning and computer vision.

---

✍️ Interactive GUI

In addition to training and evaluating the model, the project includes an interactive GUI where you can draw digits and have them recognized by the trained CNN.

---

🧱 Model Architecture

The network typically consists of:

• Conv2D layer → ReLU activation
• MaxPooling2D layer
• (Repeated stack of Conv + Pool)
• Flatten → Dense → Softmax output for 10 classes

This setup achieves robust performance (often 98–99% accuracy on test set) :contentReference[oaicite:0]{index=0}.

---

🛠️ How to Use

Requirements

pip install tensorflow numpy matplotlib

---

Training & Evaluation

python main.py

---

This will:

1. Download and preprocess MNIST dataset

2. Train the CNN model

3. Evaluate test accuracy and print results

---

📊 Results

Sample output when evaluation completes:

Test Accuracy: 0.99

(Actual numbers depend on training parameters, epochs, and TensorFlow version.)

---

📝 Optional Enhancements

You might consider extending this project with:

• Data augmentation (rotations, shifts)

• Additional Conv + Dense layers for deeper architectures

• Early stopping or checkpoint saving

• Visualizing training history using matplotlib

• Porting to TensorFlow 2.x + Keras or PyTorch

• Deploying as a web API or mobile app

📚 Learn More

Deep dive tutorial:

• Machine Learning Mastery – Building a CNN for MNIST