MNIST Digit Classification with Convolutional Neural Networks

This project aims to compare the performance of hand-written digit classification Convolutional Neural Network (CNN) models in parallel and sequential modes. Two scripts, classify_parallel.py and classify_sequential.py, are provided for this purpose. The CNN model itself is implemented in the CNN_MNIST.py script.

Getting Started

Prerequisites

Make sure you have the following dependencies installed:

• Python (>=3.6)
• NumPy
• PyTorch
• TorchVision
• Matplotlib

Installation

1. Clone this repository to your local machine using:

   git clone https://github.com/kcrmin/CNN_Hardware_Comparison.git

2. Install the required Python packages:

   pip install -r requirements.txt

Usage

1. Run the program in sequential mode:

   python classify_sequential.py

2. Run the program in parallel mode:

   python classify_parallel.py

Code Structure

CNN_MNIST.py

• Defines a CNN model for MNIST digit classification.
• Provides functions for data loading and model initialization.

classify_parallel.py

• Demonstrates parallel processing for predicting MNIST digits using the CNN model defined in CNN_MNIST.py.
• Utilizes the multiprocessing module to parallelize the prediction process, improving efficiency.

classify_sequential.py

• Demonstrates sequential processing for predicting MNIST digits using the CNN model defined in CNN_MNIST.py.
• Does not use parallelization and processes predictions sequentially.

File Structure

mnist-cnn.pth

• Pre-trained model weights saved in PyTorch format.
• Although model was trained using CNN_MNIST, had to delete the training functions due to the readability.

requirements.txt

• Contains the required Python packages to run the scripts.
• Enhance the usability as it simplifies the setup process.

Configuration

Before running the scripts, you can configure the meta_data to adjust the parameters according to your requirements. Here's a brief overview of the meta_data configuration:

   # Define meta_data
   meta_data["batch_size"] = 1
   meta_data["num_rows"] = 200
   meta_data["num_columns"] = 300
   meta_data["num_cells"] = meta_data["num_rows"] * meta_data["num_columns"]
   meta_data["threads"] = 2

Results

After running each script, the following information will be printed:

• Number of threads used (only applicable for parallel mode)
• Total number of items classified
• Total runtime
• Accuracy

Additionally, the visualization of the mask image representing the classification results will be displayed.

Screenshots

Sequential

Parallel