Fashion MNIST Classification with PyTorch CNN

Overview

This project demonstrates my proficiency in Deep Learning using PyTorch by building and training a Convolutional Neural Network (CNN) to classify images from the Fashion MNIST dataset. The Fashion MNIST dataset is a popular benchmark in machine learning, consisting of grayscale images of 10 different fashion items. This repository contains the complete code for data loading, model definition, training, and evaluation, designed to run seamlessly on Google Colab.

This project was undertaken to showcase my skills in:

• PyTorch: Utilizing PyTorch for tensor manipulation, dataset handling, model building, and training loops.
• Convolutional Neural Networks (CNNs): Designing and implementing CNN architectures suitable for image classification tasks.
• Deep Learning Fundamentals: Understanding and applying concepts like forward and backward propagation, loss functions, and optimizers.
• Data Handling: Loading and preprocessing image data using torchvision.

Author

Muyiwa Obadara (mobadara)

Table of Contents

• Overview
• Author
• Table of Contents
• Dataset
• Project Structure
• Getting Started
  • Prerequisites
  • Setup in Google Colab
• Usage
• Model Architecture
• Results (Will be updated after training)
• Future Work
• License
• Acknowledgments

Dataset

The Fashion MNIST dataset is used in this project. It contains 70,000 grayscale images of fashion products from 10 categories, with 7,000 images per category. The dataset is split into a training set of 60,000 images and a test set of 10,000 images. The 10 classes are:

1. T-shirt/top
2. Trouser
3. Pullover
4. Dress
5. Coat
6. Sandal
7. Shirt
8. Sneaker
9. Bag
10. Ankle boot

Project Structure

For Google Colab, the project structure will be simpler:

|── models
├── assets/
│   └── fashion-mnist-example.png     # Example image
├── fashion-mnist-cnn.ipynb           # Main Jupyter Notebook containing the code                 # (Optional) Script for evaluating the trained model
├── README.md                         # This file
└── requirements.txt

You can either have everything in a single notebook (fashion_mnist_cnn.ipynb) or split the model, training, and evaluation logic into separate .py files that can be imported into the notebook.

Getting Started

Prerequisites

• A Google account (to use Google Colab).
• Basic familiarity with Python.

Setup in Google Colab

1. Open a new Google Colab notebook.
2. You can either:
   • Upload all the project files (including the notebook, Python scripts if you create them, the assets folder, and requirements.txt) to the Colab environment.
   • Alternatively, you can clone this GitHub repository directly into your Colab environment using:

     !git clone [https://github.com/mobadara/fashion-mnist-pytorch-cnn.git](https://github.com/mobadara/fashion-mnist-pytorch-cnn.git)
     cd fashion-mnist-pytorch-cnn

3. If you have a requirements.txt file, you can install the necessary libraries by running:

   !pip install -r requirements.txt

Usage

Simply open the notebook in Google Colab or Kaggle and run the cells sequentially.

Model Architecture

(A detailed description of the CNN architecture you implemented will go here. For example, based on the Net class above:)

The CNN architecture consists of:

A 2D convolutional layer with 1 input channel (grayscale), 32 output channels, a kernel size of 3, and padding of 1, followed by ReLU activation. A max-pooling layer with a kernel size of 2 and a stride of 2. Another 2D convolutional layer with 32 input channels, 64 output channels, a kernel size of 3, and padding of 1, followed by ReLU activation. Another max-pooling layer with a kernel size of 2 and a stride of 2. A fully connected (linear) layer mapping the flattened features (64 * 7 * 7) to 128 output units, followed by ReLU activation. A final fully connected (linear) layer mapping 128 input units to 10 output units (one for each class).

Results

The developed Convolutional Neural Network achieved a strong performance on the Fashion MNIST dataset, attaining a test accuracy of 88.56%, with a weighted-average precision of 88.66%, and a weighted-average recall of 88.56%. Evaluation on the held-out test set demonstrated the model's ability to accurately classify unseen fashion items, with 8856 out of 10000 samples predicted correctly. Further analysis through a confusion matrix and visualization of predictions highlighted the model's overall robustness, while also revealing some confusion between specific categories like certain shirt and shoe types, providing insights for potential future improvements.

Future Work

• Experiment with different CNN architectures (e.g., adding more layers, using different activation functions).
• Implement techniques to improve performance, such as data augmentation and dropout.
• Visualise the learned filters of the convolutional layers.

License

This project is licensed under the MIT License.

Acknowledgments

The Fashion MNIST dataset provided by Zalando Research. The PyTorch @pytorch library for providing the necessary tools for building and training neural networks.