MNIST Hand-Drawn Digit Recognition

Welcome to my MNIST Hand-Drawn Digit Recognition project. This project focuses on training a deep learning model on the MNIST dataset and provides a user-friendly interface for real-time digit recognition.

Project Structure

• Jupyter Notebooks:

  • MNIST_Training_Notebook.ipynb: This notebook contains the entire process of training the model on the MNIST dataset.
  • MNIST_EDA_Notebook.ipynb: Use this notebook for an in-depth exploratory data analysis of the MNIST dataset.

• Python Scripts:

  • train_main.py: The main script for training the model on the MNIST dataset.
  • gui.py: This script launches the tkinter-based GUI for real-time digit recognition and contains functions to process and recognize hand-drawn digits.
  • model.py: Defines the architecture and parameters of the neural network model.
  • visualization.py: Utilities for visualizing dataset samples, training results, and more.
  • training.py: Contains utilities and helper functions for model training.
  • data_utils.py: Helpful utilities for loading and processing the MNIST dataset.

Getting Started

1. Ensure you have all the necessary libraries installed.

   pip install requirements.txt
   

2. Ensure you have training data MNIST in ./data folder.

3. Run the train_main.py script to train the model:

   python train_main.py
   

Running Inference with the tkinter App

The application provides an intuitive interface for users to draw digits and instantly get predictions from the trained model.

Steps:

1. Ensure the trained model is saved in the correct directory or specify the path to the trained model (e.g., ./saved_models/mnist_model.pth).

2. Start the tkinter app by running the following command in your terminal or command prompt:

   python gui.py
   

3. Once the application window opens, use your mouse to draw a digit on the canvas.

4. Click on the "Predict" button. The application will process the drawn digit and display the predicted value.

5. To clear the canvas and draw a new digit, click on the "Clear" button.

---