mp-balaji/cifar-10-image-classification-using-pytorch

An image classification project implementing a custom neural network architecture using PyTorch to accurately classify images from the CIFAR-10 dataset

CIFAR-10-Image-Classification-using-PyTorch

An image classification project implementing a custom neural network architecture using PyTorch to accurately classify images from the CIFAR-10 dataset

Overview

This project involves the implementation of a custom neural network architecture to classify images from the CIFAR-10 dataset, which consists of 60,000 32x32 color images in 10 classes, with 6,000 images per class. The dataset is split into 50,000 training images and 10,000 testing images. The repository includes the PyTorch code and a detailed summary report, describing the project architecture, findings, and results.

Repository Contents

• Code: PyTorch Implementation
• Report: Project Summary Report

Dataset

The CIFAR-10 dataset used in this project is loaded into PyTorch DataLoaders to facilitate efficient training and testing procedures. The dataset can be explored more at CIFAR-10 dataset.

Model Architecture

The neural network architecture is designed with multiple convolutional layers arranged in blocks (B1, B2, ..., BK) followed by an output block that generates logits for classification. Each block processes the input image and outputs a transformed image, which is passed to the next block or the output block.

Intermediate Blocks

• Each intermediate block comprises multiple independent convolutional layers.
• Each convolutional layer applies a set of transformations and combines their outputs weighted by a computed vector.

Output Block

• The final block computes a logits vector from the last intermediate block's output, which is used to classify the image.

Training and Evaluation

• The model is trained using cross-entropy loss.
• A batch size and various hyperparameters were set to optimize training.
• Training involves calculating loss and accuracy after each epoch, and testing is done to evaluate the model's performance.

Results

• Training loss and accuracy plots are included to demonstrate the learning process.
• The model achieved a maximum training accuracy of 95.98% and a maximum testing accuracy of 88.51% on the CIFAR-10 test set.

Techniques Used

• Various techniques from the coursework were employed to enhance model performance.
• Hyperparameter tuning was crucial in improving the accuracy of the model.

How to Run

To run this project, follow these steps:

1. Clone the repository:

   git clone https://github.com/mp-balaji/CIFAR-10-Image-Classification-using-PyTorch.git
   

2. Run the Jupyter notebook:

   jupyter notebook PyTorch_Code.ipynb