CIFAR-10 Image Classification with Fastai

Table of Contents

• Introduction
• Installation
• Usage
  • Extracting the Dataset
  • Load the Dataset
  • Training the Model
  • Making Predictions
• Model Comparison
• Acknowledgments

Introduction

This project classifies images from the CIFAR-10 dataset using various models, including a CNN, ResNet50, and a fine-tuned ResNet152. The CIFAR-10 dataset consists of 60,000 32x32 color images across 10 different classes.

Installation

To install and run this project, follow these steps:

1. Clone the repository:

   git clone https://github.com/Amnaikram1/cifar10-image-classification.git
   cd cifar10-image-classification

Usage

Extracting the Dataset

To extract the training and test datasets, use the 7z archives:

• extract_train.py: Extracts the training dataset using the py7zr package.
• extract_test.py: Extracts the test dataset using the py7zr package.

Load the Dataset

Data is loaded using the Data Loader script.

Data Loader

A data loader specifically refers to an object that encapsulates the training, validation, and optionally test datasets in a format suitable for training deep learning models. It provides a high-level interface to specify:

• Data Source
• Data Augmentation
• Batching
• Normalization

Training the Model

Train the model using Fastai by running the CNN learner script. This script defines the Image Data Loaders, creates a learner with a pre-trained ResNet152 model, and fine-tunes it.

Making Predictions

To make predictions on the test set, use the get_preds() method of CNN learner. This function will load the trained model and generate predictions for the test images.

Model Comparison

The accuracy of different models used in this project:

Model	Accuracy (%)
CNN	28%
ResNet50	46%
ResNet152 (Fine-tuned)	97%

Acknowledgments

• The Fastai library and its contributors
• Kaggle for providing the CIFAR-10 dataset
• The open-source community for their invaluable contributions

Feel free to adjust further as needed!