🚀 PyTorch MNIST Digit Classifier with Grad-CAM

This project is a Convolutional Neural Network (CNN) built in PyTorch that classifies handwritten digits (0-9). It is trained on the 60,000-image MNIST dataset and achieves high accuracy.

The key feature is the inclusion of Grad-CAM (Gradient-weighted Class Activation Mapping), which generates a visual heatmap to explain why the model made its decision. The model also provides a confidence percentage for each prediction.

✨ Features

• High-Accuracy Model: The CNN is trained for 10 epochs to achieve >98% accuracy on the test set.
• Explainable AI (XAI): Uses Grad-CAM to create a heatmap, highlighting the pixels the model "looked at" to make its prediction.
• Confidence Score: Outputs the prediction's confidence as a percentage.
• Robust Prediction: Includes a pre-processing pipeline in Cell 6 to clean, center, and format user-uploaded images for the model.

📸 Demo

Here is the model correctly identifying a user-uploaded image of the digit "2" with 99.99% confidence. The Grad-CAM heatmap clearly shows the model focused on the correct strokes of the digit.

🛠️ Technologies Used

• PyTorch
• OpenCV (cv2)
• Pillow (PIL)
• Matplotlib
• Google Colab (or Jupyter Notebook)

---

⚡ How to Use (Prediction)

This is the fast-track guide to make predictions using the included pre-trained model.

Required Files:

• MNIST_Digit_Classifier.ipynb (This notebook)
• my_mnist_model.pth (The pre-trained model file)

Steps:

1. Open the Notebook: Open the .ipynb file in Google Colab or Jupyter.
2. Upload the Model:
   • In Colab, click the folder icon (📁) on the left.
   • Click the "Upload" icon and select your my_mnist_model.pth file.
3. Run Setup Cells: Run the following cells in order:
   • Cell 1: Setup & Imports
   • Cell 2: Load Data (This defines the transform_test needed for prediction)
   • Cell 3: Model & Helper Functions (This defines the SmallCNN class)
4. SKIP Cell 4: You can skip Cell 4: Run Training entirely.
5. Load Your Model: Run Cell 5: Load Model & Grad-CAM Setup. This will load your uploaded .pth file.
6. Predict! Run Cell 6: Predict on Your Own Image and upload a picture of a digit.

---

🧠 (Optional) Train the Model From Scratch

If you don't have the .pth file or want to train the model yourself, just run all cells in order.

1. Run Cell 1: Setup & Imports
2. Run Cell 2: Load Data
3. Run Cell 3: Model & Helper Functions
4. Run Cell 4: Run Training
   • This will train the model for 10 epochs (about 2-3 minutes) and save the new my_mnist_model.pth file.
5. Run Cell 5: Load Model & Grad-CAM Setup
6. Run Cell 6: Predict on Your Own Image