ritesh-sivanathan / mnist-classification-cnn

Tensorflow/Keras convolutional neural network that is trained on MNIST number images to identify pictures of greyscale numbers.

MNIST Convolutional Neural Network (CNN)

Overview:

This project implements a Convolutional Neural Network (CNN) to classify handwritten digits from the MNIST dataset. The model is trained on a subset of the MNIST dataset consisting of 60,000 training images and tested on 12,000 test images.

Requirements:

• Python 3.10.0 ( any other version may work, try at your own discretion )
• TensorFlow 2.15.0
• Keras 2.15.0 ( separate from Tensorflow, issue on my side)
• Numpy 1.26.3

Installation:

1. Clone this repository to your local machine:

   git clone https://github.com/Ritesh-Sivanathan/mnist-classification-cnn.git

2. Navigate to the project directory:

   cd mnist-classification-cnn

3. Install the required dependencies:

   pip install -r requirements.txt

Usage:

1. Navigate to the code directory

   cd code

2. Train the model:

   python train.py

   This will train the CNN model using the MNIST training data.

3. Change to the model directory

   cd ../model

4. Evaluate the model:

   python evaluate.py

   This will evaluate the trained model on the MNIST test data and display accuracy metrics.

Results:

After training and evaluating the model in model/evalulate.py, you'll be shown the model accuracy. Feel free to tweak anything and everything to maximize the accuracy.

Some suggestions:

• Changing batch_size line 20 - train.py
• Changing filter sizes in model lines [25-34] - train.py
• Adding more image augmentation lines [21 - ...] - preprocessor.py

Model Architecture:

The CNN architecture used in this project consists of:

• Input layer (1)
• Convolutional layers (3, Conv2D)
• Max pooling layers (2, MaxPooling2D)
• Flatten layer (1)
• Dense layers (2, relu and softmax activations)
• Output layer